Add files using upload-large-folder tool

Code/Baselines/VEnhancer/.pre-commit-config.yaml

@@ -0,0 +1,24 @@
+# Exclude all third-party libraries and auto-generated files globally
+#exclude:
+repos:
+# Common hooks
+-   repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v4.6.0
+    hooks:
+    -   id: check-merge-conflict
+    -   id: check-symlinks
+    -   id: detect-private-key
+    -   id: end-of-file-fixer
+    -   id: trailing-whitespace
+        files: (.*\.(py|bzl|md|rst|c|cc|cxx|cpp|cu|h|hpp|hxx|xpu|kps|cmake|yaml|yml|hook)|BUILD|.*\.BUILD|WORKSPACE|CMakeLists\.txt)$
+# For Python files
+-   repo: https://github.com/PyCQA/isort
+    rev: 5.13.2
+    hooks:
+    -   id: isort
+-   repo: https://github.com/psf/black.git
+    rev: 24.4.2
+    hooks:
+    -   id: black
+        files: (.*\.(py|pyi|bzl)|BUILD|.*\.BUILD|WORKSPACE)$
+        args: [--line-length=120]

Code/Baselines/VEnhancer/README.md

@@ -0,0 +1,173 @@
+<div align="center">
+
+<h1>VEnhancer: Generative Space-Time Enhancement<br>for Video Generation</h1>
+
+<div>
+    <a href='https://scholar.google.com/citations?user=GUxrycUAAAAJ&hl=zh-CN' target='_blank'>Jingwen He</a>,&emsp;
+    <a href='https://tianfan.info' target='_blank'>Tianfan Xue</a>,&emsp;
+    <a href='https://github.com/ChrisLiu6' target='_blank'>Dongyang Liu</a>,&emsp;
+    <a href='https://github.com/0x3f3f3f3fun' target='_blank'>Xinqi Lin</a>,&emsp;
+</div>
+    <a href='https://gaopengcuhk.github.io' target='_blank'>Peng Gao</a>,&emsp;
+    <a href='https://scholar.google.com/citations?user=GMzzRRUAAAAJ&hl=en' target='_blank'>Dahua Lin</a>,&emsp;
+    <a href='https://scholar.google.com/citations?user=gFtI-8QAAAAJ&hl=en' target='_blank'>Yu Qiao</a>,&emsp;
+    <a href='https://wlouyang.github.io' target='_blank'>Wanli Ouyang</a>,&emsp;
+    <a href='https://liuziwei7.github.io' target='_blank'>Ziwei Liu</a>
+<div>
+</div>
+<div>
+    The Chinese University of Hong Kong,&emsp;Shanghai Artificial Intelligence Laboratory,&emsp;
+</div>
+<div>
+
+</div>
+<div>
+ S-Lab, Nanyang Technological University&emsp;
+</div>
+
+<div>
+    <h4 align="center">
+        <a href="https://vchitect.github.io/VEnhancer-project/" target='_blank'>
+        <img src="https://img.shields.io/badge/🐳-Project%20Page-blue">
+        </a>
+        <a href="https://arxiv.org/abs/2407.07667" target='_blank'>
+        <img src="https://img.shields.io/badge/arXiv-2312.06640-b31b1b.svg">
+        </a>
+        <a href="https://youtu.be/QMR_5weifGg" target='_blank'>
+        <img src="https://img.shields.io/badge/Demo%20Video-%23FF0000.svg?logo=YouTube&logoColor=white">
+        <!-- </a>
+        <img src="https://api.infinitescript.com/badgen/count?name="> -->
+    </h4>
+</div>
+
+<strong>VEnhancer, an All-in-One generative video enhancement model that can achieve spatial super-resolution, temporal super-resolution, and video refinement for AI-generated videos.</strong>
+
+<table class="center">
+  <tr>
+    <td colspan="1">AIGC video</td>
+    <td colspan="1">+VEnhancer</td>
+  </tr>
+  <tr>
+  <td>
+    <img src=assets/input_fish.gif width="380">
+  </td>
+  <td>
+    <img src=assets/out_fish.gif width="380">
+  </td>
+  </tr>
+
+
+
+</table>
+
+:open_book: For more visual results, go checkout our <a href="https://vchitect.github.io/VEnhancer-project/" target="_blank">project page</a>
+
+
+---
+
+</div>
+
+
+## 🔥 Update
+- [2024.09.12] 😸 Release our version 2 checkpoint: **[venhancer_v2.pt](https://huggingface.co/jwhejwhe/VEnhancer/resolve/main/venhancer_v2.pt)** . It is less creative, but is able to generate more texture details, and has better identity preservation, which is more suitable for enhancing videos with profiles.
+- [2024.09.10] 😸 Support **Multiple GPU Inference** and **tiled VAE** for temporal VAE decoding. And more stable performance for long video enhancement.
+- [2024.08.18] 😸 Support enhancement for **abitrary long videos** (by spliting the videos into muliple chunks with overlaps); **Faster sampling** with only 15 steps without obvious quality loss (by setting `--solver_mode 'fast'` in the script command); Use **temporal VAE** to reduce video flickering.
+- [2024.07.28] 🔥 Inference code and pretrained video enhancement model are released.
+- [2024.07.10] 🤗 This repo is created.
+
+<!-- ## Open Source Plan
+
+- [x] Release code of Multiple GPU Inference.
+- [x] Release code of tiled VAE.
+- [ ] Release model that is optimized for better idenity preservation. -->
+
+<!-- :star::star::star: Star us :star::star::star:! And we will speed up the open-sourcing process :heart:. -->
+
+## :astonished: Gallery
+
+
+| Inputs & Results | Model Version |
+| :---------- | :-: |
+|Prompt: A close-up shot of a woman standing in a dimly lit room. she is wearing a traditional chinese outfit, which includes a red and gold dress with intricate designs and a matching headpiece.<br/><video src="https://github.com/user-attachments/assets/4a514853-65f6-40b8-8b5d-d14835bb9297" width="100%" controls autoplay></video>from [Open-Sora](https://github.com/hpcaitech/Open-Sora)|<div style="width:100px">v2</div>|
+|Prompt: Einstein plays guitar.<br/><table class="center"><tr><td><video src="https://github.com/user-attachments/assets/aa76e8a2-14e2-49a1-915c-147838476ab1" width="50%" controls autoplay></video></td><td><video src="https://github.com/user-attachments/assets/f08e6f77-19d4-4847-9356-739a84da38b2" width="50%" controls autoplay></video></td></tr></table>from [Kling](https://kling.kuaishou.com/en)|<div style="width:100px">v2</div>|
+|Prompt: A girl eating noodles.<br/><table class="center"><tr><td><video src="https://github.com/user-attachments/assets/cc01bf80-8b49-4314-97a3-1e1ec2d16d6a" width="50%" controls autoplay></video></td><td><video src="https://github.com/user-attachments/assets/ce923609-614b-4f87-ba2b-7b831edce40f" width="50%" controls autoplay></video></td></tr></table>from [Kling](https://kling.kuaishou.com/en)| <div style="width:100px">v2</div>|
+|Prompt: A little brick man visiting an art gallery.<br/><video src="https://github.com/user-attachments/assets/39a39459-4a69-4ef7-80ef-74df066decb5" width="100%" controls autoplay></video><br/><video src="https://github.com/user-attachments/assets/d110bec4-9ea1-4348-a6db-e9dd6cce4bc2" width="100%" controls autoplay></video>from [Kling](https://kling.kuaishou.com/en) | <div style="width:100px">v1</div>|
+<!-- |Prompt: A detailed wooden toy ship with intricately carved masts and sails is seen gliding smoothly over a plush, blue carpet that mimics the waves of the sea.<br/><video src="https://github.com/user-attachments/assets/d6ba4ebe-a970-4db1-ade1-03bfa8e52a20" width="100%" controls autoplay></video><video src="https://github.com/user-attachments/assets/bf97116e-2fbc-4e29-b559-4fe08dc65c02" width="100%" controls autoplay></video>from [CogVideoX](https://github.com/THUDM/CogVideo)|<div style="width:100px">v2</div>| -->
+
+## 🎬 Overview
+VEnhancer achieves spatial super-resolution, temporal super-resolution (i.e, frame interpolation), and video refinement in **one model**.
+It is flexible to adapt to different upsampling factors (e.g., 1x~8x) for either spatial or temporal super-resolution. Besides, it provides flexible control to modify the refinement strength for handling diversified video artifacts.
+
+It follows ControlNet and copies the architecures and weights of multi-frame encoder and middle block of a pretrained video diffusion model to build a trainable condition network.
+This **video ControlNet** accepts both low-resolution key frames and full frames of noisy latents as inputs.
+Also, the noise level $\sigma$ regarding noise augmentation and downscaling factor $s$ serve as additional network conditioning through our proposed **video-aware conditioning** apart from timestep $t$ and prompt $c_{text}$.
+<!-- ![overall_structure](assets/venhancer_arch.png) -->
+
+
+## :gear: Installation
+```shell
+# clone this repo
+git clone https://github.com/Vchitect/VEnhancer.git
+cd VEnhancer
+
+# create environment
+conda create -n venhancer python=3.10
+conda activate venhancer
+pip install torch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2
+pip install -r requirements.txt
+```
+Note that ffmpeg command should be enabled. If you have sudo access, then you can install it using the following command:
+```shell
+sudo apt-get update && apt-get install ffmpeg libsm6 libxext6  -y
+```
+
+## :dna: Pretrained Models
+| Model Name | Description | HuggingFace | BaiduNetdisk  |
+| :---------: | :----------: | :----------: | :----------: |
+| venhancer_paper.pth  | very creative, strong refinement, but sometimes over-smooths edges and texture details. | [download](https://huggingface.co/jwhejwhe/VEnhancer/resolve/main/venhancer_paper.pt?download=true) | [download](https://pan.baidu.com/s/15t20RGvEHqJOMmhA_zRLiA?pwd=cpsd)|
+| venhancer_v2.pth  | less creative, but can generate better texture details, and has better identity preservation. | [download](https://huggingface.co/jwhejwhe/VEnhancer/resolve/main/venhancer_v2.pt?download=true) | [download](https://pan.baidu.com/s/1mc4s5xqcVqKyL-GwkE0loA?pwd=bbqn)|
+
+## 💫 Inference
+1) Download the VEnhancer model and then put the checkpoint in the `VEnhancer/ckpts` directory. (optional as it can be done automatically)
+2) run the following command.
+```bash
+  bash run_VEnhancer.sh
+```
+for single GPU inference (at least A100 80G is required), or
+```bash
+  bash run_VEnhancer_MultiGPU.sh
+```
+for multiple GPU inference.
+
+In `run_VEnhancer.sh` or `run_VEnhancer_MultiGPU.sh`,
+- `version`. We now provide two choices: `v1` and `v2` (venhancer_paper.pth and venhancer_v2.pth, respectively).
+- `up_scale` is the upsampling factor ($1\sim8$) for spatial super-resolution. $\times3,4$ are recommended. Note that the target resolution will be adjusted no higher than 2k resolution.
+- `target_fps` is your expected target fps, and the default is 24.
+- `noise_aug` is the noise level ($0\sim300$) regarding noise augmentation. Higher noise corresponds to stronger refinement. $200\sim300$ are recommended.
+- Regarding prompt, you can use `--filename_as_prompt` to automatically use filename as prompt; or you can write the prompt to a txt file, and specify the prompt_path by setting `--prompt_path [your_prompt_path]`; or directly provide the prompt by specifying `--prompt [your_prompt]`.
+- Regarding sampling, `--solver_mode fast` has fixed 15 sampling steps. For `--solver_mode normal`, you can modify `steps` to trade off efficiency over video quality.
+
+### Gradio
+The same functionality is also available as a gradio demo. Please follow the previous guidelines, and specify the model version (v1 or v2).
+``` shell
+python gradio_app.py --version v1
+```
+
+
+## BibTeX
+If you use our work in your research, please cite our publication:
+```
+@article{he2024venhancer,
+  title={VEnhancer: Generative Space-Time Enhancement for Video Generation},
+  author={He, Jingwen and Xue, Tianfan and Liu, Dongyang and Lin, Xinqi and Gao, Peng and Lin, Dahua and Qiao, Yu and Ouyang, Wanli and Liu, Ziwei},
+  journal={arXiv preprint arXiv:2407.07667},
+  year={2024}
+}
+```
+
+## 🤗 Acknowledgements
+Our codebase builds on [modelscope](https://github.com/modelscope/modelscope).
+Thanks the authors for sharing their awesome codebases!
+
+## 📧 Contact
+If you have any questions, please feel free to reach us at `hejingwenhejingwen@outlook.com`.

Code/Baselines/VEnhancer/enhance_a_video_MultiGPU.py

@@ -0,0 +1,215 @@
+from argparse import ArgumentParser, Namespace
+import glob
+
+from easydict import EasyDict
+from huggingface_hub import hf_hub_download
+import torch.cuda
+import torch.distributed as dist
+
+from inference_utils import *
+from video_to_video.context_parallel import get_context_parallel_rank, initialize_context_parallel
+from video_to_video.utils.seed import setup_seed
+from video_to_video.video_to_video_model_parallel import VideoToVideoParallel
+
+logger = get_logger()
+
+
+class VEnhancer:
+    def __init__(
+        self,
+        result_dir="./results/",
+        version="v1",
+        model_path="",
+        solver_mode="fast",
+        steps=15,
+        guide_scale=7.5,
+        s_cond=8,
+    ):
+        if not model_path:
+            self.download_model(version=version)
+        else:
+            self.model_path = model_path
+        assert os.path.exists(self.model_path), "Error: checkpoint Not Found!"
+        logger.info(f"checkpoint_path: {self.model_path}")
+
+        self.result_dir = result_dir
+        os.makedirs(self.result_dir, exist_ok=True)
+
+        model_cfg = EasyDict(__name__="model_cfg")
+        model_cfg.model_path = self.model_path
+        self.model = VideoToVideoParallel(model_cfg)
+
+        steps = 15 if solver_mode == "fast" else steps
+        self.solver_mode = solver_mode
+        self.steps = steps
+        self.guide_scale = guide_scale
+        self.s_cond = s_cond
+
+    def enhance_a_video(self, video_path, prompt, up_scale=4, target_fps=24, noise_aug=300):
+
+        save_name = os.path.splitext(os.path.basename(video_path))[0]
+        text = prompt
+        logger.info(f"text: {text}")
+        caption = text + self.model.positive_prompt
+
+        input_frames, input_fps = load_video(video_path)
+        in_f_num = len(input_frames)
+        logger.info(f"input frames length: {in_f_num}")
+        logger.info(f"input fps: {input_fps}")
+        interp_f_num = max(round(target_fps / input_fps) - 1, 0)
+        interp_f_num = min(interp_f_num, 8)
+        target_fps = input_fps * (interp_f_num + 1)
+        logger.info(f"target_fps: {target_fps}")
+
+        video_data = preprocess(input_frames)
+        _, _, h, w = video_data.shape
+        logger.info(f"input resolution: {(h, w)}")
+        target_h, target_w = adjust_resolution(h, w, up_scale)
+        logger.info(f"target resolution: {(target_h, target_w)}")
+
+        mask_cond = make_mask_cond(in_f_num, interp_f_num)
+        mask_cond = torch.Tensor(mask_cond).long()
+
+        noise_aug = min(max(noise_aug, 0), 300)
+        logger.info(f"noise augmentation: {noise_aug}")
+        logger.info(f"scale s is set to: {self.s_cond}")
+
+        pre_data = {"video_data": video_data, "y": caption}
+        pre_data["mask_cond"] = mask_cond
+        pre_data["s_cond"] = self.s_cond
+        pre_data["interp_f_num"] = interp_f_num
+        pre_data["target_res"] = (target_h, target_w)
+        pre_data["t_hint"] = noise_aug
+
+        total_noise_levels = 900
+        setup_seed(666)
+
+        with torch.no_grad():
+            data_tensor = collate_fn(pre_data, "cuda")
+            output = self.model.test(
+                data_tensor,
+                total_noise_levels,
+                steps=self.steps,
+                solver_mode=self.solver_mode,
+                guide_scale=self.guide_scale,
+                noise_aug=noise_aug,
+            )
+
+        output = tensor2vid(output)
+
+        if get_context_parallel_rank() == 0:
+            save_video(output, self.result_dir, f"{save_name}.mp4", fps=target_fps)
+        dist.barrier()
+        return os.path.join(self.result_dir, save_name)
+
+    def download_model(self, version):
+        REPO_ID = "jwhejwhe/VEnhancer"
+        filename = "venhancer_paper.pt"
+        if version == "v2":
+            filename = "venhancer_v2.pt"
+        ckpt_dir = "./ckpts/"
+        os.makedirs(ckpt_dir, exist_ok=True)
+        local_file = os.path.join(ckpt_dir, filename)
+        if not os.path.exists(local_file):
+            logger.info(f"Downloading the VEnhancer checkpoint...")
+            hf_hub_download(repo_id=REPO_ID, filename=filename, local_dir=ckpt_dir)
+        self.model_path = local_file
+
+
+def parse_args() -> Namespace:
+    parser = ArgumentParser()
+
+    parser.add_argument("--input_path", required=True, type=str, help="input video path")
+    parser.add_argument("--save_dir", type=str, default="results", help="save directory")
+    parser.add_argument("--version", type=str, default="v1", choices=["v1", "v2"], help="model version")
+    parser.add_argument("--model_path", type=str, default="", help="model path")
+
+    parser.add_argument("--prompt", type=str, default="a good video", help="prompt")
+    parser.add_argument("--prompt_path", type=str, default="", help="prompt path")
+    parser.add_argument("--filename_as_prompt", action="store_true")
+
+    parser.add_argument("--cfg", type=float, default=7.5)
+    parser.add_argument("--solver_mode", type=str, default="fast", choices=["fast", "normal"], help="fast | normal")
+    parser.add_argument("--steps", type=int, default=15)
+
+    parser.add_argument("--noise_aug", type=int, default=200, help="noise augmentation")
+    parser.add_argument("--target_fps", type=int, default=24)
+    parser.add_argument("--up_scale", type=float, default=4)
+    parser.add_argument("--s_cond", type=float, default=8)
+
+    return parser.parse_args()
+
+
+def main():
+
+    args = parse_args()
+
+    world_size = int(os.environ["WORLD_SIZE"])
+    rank = int(os.environ["RANK"])
+    gpu_id = int(os.environ["LOCAL_RANK"])
+
+    dist.init_process_group(
+        backend="nccl",
+        rank=rank,
+        world_size=world_size,
+        init_method="env://",
+    )
+    torch.cuda.set_device(gpu_id)
+
+    initialize_context_parallel(world_size)
+
+    input_path = args.input_path
+    prompt = args.prompt
+    prompt_path = args.prompt_path
+    filename_as_prompt = args.filename_as_prompt
+    model_path = args.model_path
+    version = args.version
+    save_dir = args.save_dir
+
+    noise_aug = args.noise_aug
+    up_scale = args.up_scale
+    target_fps = args.target_fps
+    s_cond = args.s_cond
+
+    steps = args.steps
+    solver_mode = args.solver_mode
+    guide_scale = args.cfg
+
+    venhancer = VEnhancer(
+        result_dir=save_dir,
+        version=version,
+        model_path=model_path,
+        solver_mode=solver_mode,
+        steps=steps,
+        guide_scale=guide_scale,
+        s_cond=s_cond,
+    )
+
+    if os.path.isdir(input_path):
+        file_path_list = sorted(glob.glob(os.path.join(input_path, "*.mp4")))
+    elif os.path.isfile(input_path):
+        file_path_list = [input_path]
+    else:
+        raise TypeError("input must be a directory or video file!")
+
+    prompt_list = None
+    if os.path.isfile(prompt_path):
+        prompt_list = load_prompt_list(prompt_path)
+        assert len(prompt_list) == len(file_path_list)
+
+    for ind, file_path in enumerate(file_path_list):
+        logger.info(f"processing video {ind}, file_path: {file_path}")
+        if filename_as_prompt:
+            prompt = os.path.splitext(os.path.basename(file_path))[0]
+        elif prompt_list is not None:
+            prompt = prompt_list[ind]
+        else:
+            prompt_path = os.path.splitext(file_path)[0] + ".txt"
+            if os.path.isfile(prompt_path):
+                logger.info(f"prompt_path: {prompt_path}")
+                prompt = load_prompt_list(prompt_path)[0]
+        venhancer.enhance_a_video(file_path, prompt, up_scale, target_fps, noise_aug)
+
+
+if __name__ == "__main__":
+    main()

Code/Baselines/VEnhancer/gradio_app.py

@@ -0,0 +1,88 @@
+import argparse
+import os
+import random
+
+import gradio as gr
+
+from enhance_a_video import VEnhancer
+
+examples = [
+    ["prompts/astronaut.mp4", "An astronaut flying in space, featuring a steady and smooth perspective", 4, 24, 250],
+    ["prompts/cat.mp4", "A cat wearing sunglasses at a pool", 4, 24, 250],
+    ["prompts/fish.mp4", "Clown fish swimming through the coral reef", 4, 24, 300],
+    ["prompts/iron_man.mp4", "Iron Man flying in the sky", 4, 24, 250],
+    ["prompts/raccoon.mp4", "A cute raccoon playing guitar in a boat on the ocean", 4, 24, 250],
+    ["prompts/shanghai.mp4", "The bund Shanghai by Hokusai, in the style of Ukiyo", 4, 24, 250],
+    ["prompts/gwen.mp4", "Gwen Stacy reading a book, black and white", 4, 24, 250],
+]
+
+
+def venhancer_demo(result_dir="./tmp/", version="v1"):
+    css = """#input_video {max-width: 1024px !important} #output_vid {max-width: 2048px; max-height:1280px}"""
+    venhancer = VEnhancer(result_dir, version=version)
+    with gr.Blocks(analytics_enabled=False, css=css) as venhancer_iface:
+        gr.Markdown(
+            "<div align='center'> <h1> VEnhancer </span> </h1> \
+                     <a style='font-size:18px;color: #000000' href='https://arxiv.org/abs/2407.07667'> [ArXiv] </a>\
+                     <a style='font-size:18px;color: #000000' href='https://vchitect.github.io/VEnhancer-project/'> [Project Page] </a> \
+                     <a style='font-size:18px;color: #000000' href='https://github.com/Vchitect/VEnhancer'> [Github] </a> </div>"
+        )
+        with gr.Tab(label="VEnhancer"):
+            with gr.Column():
+                with gr.Row():
+                    with gr.Column():
+                        with gr.Row():
+                            input_video = gr.Video(label="Input Video", elem_id="input_video")
+                        with gr.Row():
+                            input_text = gr.Text(label="Prompts")
+                        with gr.Row():
+                            up_scale = gr.Slider(
+                                minimum=1.0, maximum=8.0, step=0.1, label="up scale", value=4, elem_id="up_scale"
+                            )
+                        with gr.Row():
+                            target_fps = gr.Slider(
+                                minimum=8, maximum=60, step=1, elem_id="target_fps", label="target fps", value=24
+                            )
+                        with gr.Row():
+                            noise_aug = gr.Slider(
+                                minimum=50, maximum=300, step=1, elem_id="noise_aug", label="noise aug", value=250
+                            )
+                        end_btn = gr.Button("Generate")
+                    with gr.Row():
+                        output_video = gr.Video(
+                            label="Generated Video", elem_id="output_vid", autoplay=True, show_share_button=True
+                        )
+
+                gr.Examples(
+                    examples=examples,
+                    inputs=[input_video, input_text, up_scale, target_fps, noise_aug],
+                    outputs=[output_video],
+                    fn=venhancer.enhance_a_video,
+                    cache_examples=False,
+                )
+            end_btn.click(
+                inputs=[input_video, input_text, up_scale, target_fps, noise_aug],
+                outputs=[output_video],
+                fn=venhancer.enhance_a_video,
+            )
+
+    return venhancer_iface
+
+
+def get_parser():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--version", type=str, default="v1", choices=["v1", "v2"], help="select the checkpoint version."
+    )
+
+    return parser
+
+
+if __name__ == "__main__":
+    parser = get_parser()
+    args = parser.parse_args()
+
+    result_dir = os.path.join("./", "results")
+    venhancer_iface = venhancer_demo(result_dir, args.version)
+    venhancer_iface.queue(max_size=12)
+    venhancer_iface.launch(max_threads=1)

Code/Baselines/VEnhancer/inference_utils.py

@@ -0,0 +1,153 @@
+import os
+import subprocess
+import tempfile
+from typing import Mapping
+
+from PIL import Image
+import cv2
+from einops import rearrange
+import numpy as np
+import torch
+import torch.nn.functional as F
+import torchvision.transforms.functional as transforms_F
+
+from video_to_video.utils.logger import get_logger
+
+logger = get_logger()
+
+
+def tensor2vid(video, mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5]):
+    mean = torch.tensor(mean, device=video.device).reshape(1, -1, 1, 1, 1)
+    std = torch.tensor(std, device=video.device).reshape(1, -1, 1, 1, 1)
+    video = video.mul_(std).add_(mean)
+    video.clamp_(0, 1)
+    video = video * 255.0
+    images = rearrange(video, "b c f h w -> b f h w c")[0]
+    return images
+
+
+def preprocess(input_frames):
+    out_frame_list = []
+    for pointer in range(len(input_frames)):
+        frame = input_frames[pointer]
+        frame = frame[:, :, ::-1]
+        frame = Image.fromarray(frame.astype("uint8")).convert("RGB")
+        frame = transforms_F.to_tensor(frame)
+        out_frame_list.append(frame)
+    out_frames = torch.stack(out_frame_list, dim=0)
+    out_frames.clamp_(0, 1)
+    mean = out_frames.new_tensor([0.5, 0.5, 0.5]).view(-1)
+    std = out_frames.new_tensor([0.5, 0.5, 0.5]).view(-1)
+    out_frames.sub_(mean.view(1, -1, 1, 1)).div_(std.view(1, -1, 1, 1))
+    return out_frames
+
+
+def adjust_resolution(h, w, up_scale):
+    if h * w * up_scale * up_scale < 720 * 1280 * 1.5:
+        up_s = np.sqrt(720 * 1280 * 1.5 / (h * w))
+        target_h = int(up_s * h // 2 * 2)
+        target_w = int(up_s * w // 2 * 2)
+    elif h * w * up_scale * up_scale > 1152 * 2048:
+        up_s = np.sqrt(1152 * 2048 / (h * w))
+        target_h = int(up_s * h // 2 * 2)
+        target_w = int(up_s * w // 2 * 2)
+    else:
+        target_h = int(up_scale * h // 2 * 2)
+        target_w = int(up_scale * w // 2 * 2)
+    return (target_h, target_w)
+
+
+def make_mask_cond(in_f_num, interp_f_num):
+    mask_cond = []
+    interp_cond = [-1 for _ in range(interp_f_num)]
+    for i in range(in_f_num):
+        mask_cond.append(i)
+        if i != in_f_num - 1:
+            mask_cond += interp_cond
+    return mask_cond
+
+
+def load_prompt_list(file_path):
+    files = []
+    with open(file_path, "r") as fin:
+        for line in fin:
+            path = line.strip()
+            if path:
+                files.append(path)
+    return files
+
+
+def load_video(vid_path):
+    capture = cv2.VideoCapture(vid_path)
+    _fps = capture.get(cv2.CAP_PROP_FPS)
+    _total_frame_num = capture.get(cv2.CAP_PROP_FRAME_COUNT)
+    pointer = 0
+    frame_list = []
+    stride = 1
+    while len(frame_list) < _total_frame_num:
+        ret, frame = capture.read()
+        pointer += 1
+        if (not ret) or (frame is None):
+            break
+        if pointer >= _total_frame_num + 1:
+            break
+        if pointer % stride == 0:
+            frame_list.append(frame)
+    capture.release()
+    return frame_list, _fps
+
+
+def save_video(video, save_dir, file_name, fps=16.0):
+    output_path = os.path.join(save_dir, file_name)
+    images = [(img.numpy()).astype("uint8") for img in video]
+    temp_dir = tempfile.mkdtemp()
+    for fid, frame in enumerate(images):
+        tpth = os.path.join(temp_dir, "%06d.png" % (fid + 1))
+        cv2.imwrite(tpth, frame[:, :, ::-1])
+    tmp_path = os.path.join(save_dir, "tmp.mp4")
+    cmd = f"ffmpeg -y -f image2 -framerate {fps} -i {temp_dir}/%06d.png \
+     -vcodec libx264 -crf 17 -pix_fmt yuv420p {tmp_path}"
+    status, output = subprocess.getstatusoutput(cmd)
+    if status != 0:
+        logger.error(f"Save Video Error with {output}")
+    os.system(f"rm -rf {temp_dir}")
+    os.rename(tmp_path, output_path)
+
+
+def collate_fn(data, device):
+    """Prepare the input just before the forward function.
+    This method will move the tensors to the right device.
+    Usually this method does not need to be overridden.
+
+    Args:
+        data: The data out of the dataloader.
+        device: The device to move data to.
+
+    Returns: The processed data.
+
+    """
+    from torch.utils.data.dataloader import default_collate
+
+    def get_class_name(obj):
+        return obj.__class__.__name__
+
+    if isinstance(data, dict) or isinstance(data, Mapping):
+        return type(data)({k: collate_fn(v, device) if k != "img_metas" else v for k, v in data.items()})
+    elif isinstance(data, (tuple, list)):
+        if 0 == len(data):
+            return torch.Tensor([])
+        if isinstance(data[0], (int, float)):
+            return default_collate(data).to(device)
+        else:
+            return type(data)(collate_fn(v, device) for v in data)
+    elif isinstance(data, np.ndarray):
+        if data.dtype.type is np.str_:
+            return data
+        else:
+            return collate_fn(torch.from_numpy(data), device)
+    elif isinstance(data, torch.Tensor):
+        return data.to(device)
+    elif isinstance(data, (bytes, str, int, float, bool, type(None))):
+        return data
+    else:
+        raise ValueError(f"Unsupported data type {type(data)}")

Code/Baselines/VEnhancer/requirements.txt

@@ -0,0 +1,12 @@
+opencv-python==4.10.0.84
+easydict==1.13
+einops==0.8.0
+open-clip-torch==2.20.0
+xformers==0.0.21
+fairscale==0.4.13
+torchsde==0.2.6
+pytorch-lightning==2.0.1
+diffusers==0.30.0
+huggingface_hub==0.23.3
+gradio==4.41.0
+numpy==1.24

Code/Baselines/VEnhancer/run_VEnhancer.sh

@@ -0,0 +1,7 @@
+python enhance_a_video.py \
+--version v1 \
+--up_scale 4 --target_fps 24 --noise_aug 250 \
+--solver_mode 'fast' --steps 15 \
+--input_path prompts \
+--prompt_path prompts/text_prompts.txt \
+--save_dir 'results/'

Code/Baselines/VEnhancer/run_VEnhancer_MultiGPU.sh

@@ -0,0 +1,9 @@
+N_PARALLEL_GPUS=4
+
+torchrun --nproc_per_node=${N_PARALLEL_GPUS} enhance_a_video_MultiGPU.py \
+--version v1 \
+--up_scale 4 --target_fps 24 --noise_aug 250 \
+--solver_mode 'fast' --steps 15 \
+--input_path prompts \
+--prompt_path prompts/text_prompts.txt \
+--save_dir "results_with_${N_PARALLEL_GPUS}gpu/"

Code/Baselines/VEnhancer/run_VEnhancer_MultiGPU_demo.sh

@@ -0,0 +1,9 @@
+N_PARALLEL_GPUS=8
+
+torchrun --nproc_per_node=${N_PARALLEL_GPUS} enhance_a_video_MultiGPU.py \
+--version v1 \
+--up_scale 3 --target_fps 24 --noise_aug 250 \
+--solver_mode 'fast' --steps 15 \
+--input_path prompts_demo \
+--prompt_path prompts_demo/chair_prompt.txt \
+--save_dir "results_with_${N_PARALLEL_GPUS}gpu/"

Code/Baselines/flash-attention/benchmarks/benchmark_alibi.py

@@ -0,0 +1,275 @@
+# Copyright (c) 2024, Sanghun Cho, Tri Dao.
+
+import pickle
+import math
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from einops import rearrange, repeat
+from flash_attn.layers.rotary import apply_rotary_emb
+
+from flash_attn.utils.benchmark import benchmark_all, benchmark_forward, benchmark_backward
+from flash_attn.utils.benchmark import benchmark_fwd_bwd, benchmark_combined
+
+from flash_attn import flash_attn_qkvpacked_func, flash_attn_func
+
+try:
+    import xformers.ops as xops
+except ImportError:
+    xops = None
+
+
+def generate_cos_sin(seqlen, rotary_dim, device, dtype):
+    assert rotary_dim % 2 == 0
+    angle = torch.rand(seqlen * 2, rotary_dim // 2, device=device) * 2 * math.pi
+    cos = torch.cos(angle).to(dtype=dtype)
+    sin = torch.sin(angle).to(dtype=dtype)
+    return cos, sin
+
+
+def flash_rotary(q, k, v, cos, sin, causal=False):
+    # corrected by @tridao comments
+    q = apply_rotary_emb(
+        q, cos, sin, seqlen_offsets=0, interleaved=False, inplace=True
+    )
+    k = apply_rotary_emb(
+        k, cos, sin, seqlen_offsets=0, interleaved=False, inplace=True
+    )
+
+    return flash_attn_func(q, k, v, causal=causal)
+
+
+def attn_bias_from_alibi_slopes(
+    slopes, seqlen_q, seqlen_k, query_padding_mask=None, key_padding_mask=None, causal=False
+):
+    batch, nheads = slopes.shape
+    device = slopes.device
+    slopes = rearrange(slopes, "b h -> b h 1 1")
+    if causal:
+        return torch.arange(-seqlen_k + 1, 1, device=device, dtype=torch.float32) * slopes
+    else:
+        row_idx = rearrange(torch.arange(seqlen_q, device=device, dtype=torch.long), "s -> s 1")
+        col_idx = torch.arange(seqlen_k, device=device, dtype=torch.long)
+        sk = (
+            seqlen_k
+            if key_padding_mask is None
+            else rearrange(key_padding_mask.sum(-1), "b -> b 1 1 1")
+        )
+        sq = (
+            seqlen_q
+            if query_padding_mask is None
+            else rearrange(query_padding_mask.sum(-1), "b -> b 1 1 1")
+        )
+        relative_pos = torch.abs(row_idx + sk - sq - col_idx)
+        return -slopes * relative_pos.to(dtype=slopes.dtype)
+
+
+def flops(batch, seqlen, headdim, nheads, causal, mode="fwd"):
+    assert mode in ["fwd", "bwd", "fwd_bwd"]
+    f = 4 * batch * seqlen**2 * nheads * headdim // (2 if causal else 1)
+    return f if mode == "fwd" else (2.5 * f if mode == "bwd" else 3.5 * f)
+
+
+def efficiency(flop, time):
+    return (flop / time / 10**12) if not math.isnan(time) else 0.0
+
+
+def attention_pytorch(q, k, v, dropout_p=0.0, causal=True, attn_bias=None):
+    """
+    Arguments:
+        q, k, v: (batch_size, seqlen, nheads, head_dim)
+        dropout_p: float
+        attn_bias: (batch_size, nheads, seqlen, seqlen) or (1, nheads, seqlen, seqlen)
+    Output:
+        output: (batch_size, seqlen, nheads, head_dim)
+    """
+    batch_size, seqlen, nheads, d = q.shape
+    q = rearrange(q, 'b t h d -> (b h) t d')
+    k = rearrange(k, 'b s h d -> (b h) d s')
+    softmax_scale = 1.0 / math.sqrt(d)
+    # Preallocate attn_weights for `baddbmm`
+    if attn_bias is not None:
+        scores = rearrange(attn_bias, 'b h t s -> (b h) t s')
+    else:
+        scores = torch.empty(batch_size * nheads, seqlen, seqlen, dtype=q.dtype, device=q.device)
+    scores = rearrange(torch.baddbmm(scores, q, k, beta=1.0, alpha=softmax_scale),
+                       '(b h) t s -> b h t s', h=nheads)
+    if causal:
+        # "triu_tril_cuda_template" not implemented for 'BFloat16'
+        # So we have to construct the mask in float
+        causal_mask = torch.triu(torch.full((seqlen, seqlen), -10000.0, device=scores.device), 1)
+        # TD [2022-09-30]: Adding is faster than masked_fill_ (idk why, just better kernel I guess)
+        scores = scores + causal_mask.to(dtype=scores.dtype)
+    attention = torch.softmax(scores, dim=-1)
+    attention_drop = F.dropout(attention, dropout_p)
+    output = torch.einsum('bhts,bshd->bthd', attention_drop , v)
+    return output.to(dtype=q.dtype)
+
+
+def time_fwd_bwd(func, *args, **kwargs):
+    time_f, time_b = benchmark_fwd_bwd(func, *args, **kwargs)
+    return time_f[1].mean, time_b[1].mean
+
+
+repeats = 30
+device = 'cuda'
+dtype = torch.float16
+
+bs_seqlen_vals = [(32, 512), (16, 1024), (8, 2048), (4, 4096), (2, 8192), (1, 16384)]
+causal_vals = [False, True]
+headdim_vals = [64, 128]
+dim = 2048
+dropout_p = 0.0
+
+methods = (["fa2_alibi", "torch"]
+           + (["xformers"] if xops is not None else [])
+           + ["sdpa"]
+           + ["fa2_baseline"]
+           + ["fa2_rotary"])
+
+time_f = {}
+time_b = {}
+time_f_b = {}
+speed_f = {}
+speed_b = {}
+speed_f_b = {}
+for causal in causal_vals:
+    for headdim in headdim_vals:
+        for batch_size, seqlen in bs_seqlen_vals:
+            config = (causal, headdim, batch_size, seqlen)
+            nheads = dim // headdim
+            q, k, v = [torch.randn(batch_size, seqlen, nheads, headdim, device=device, dtype=dtype,
+                                    requires_grad=True) for _ in range(3)]
+            # alibi_slopes = torch.rand(batch_size, nheads, device=device, dtype=torch.float32) * 0.3
+            alibi_slopes = torch.rand(1, nheads, device=device, dtype=torch.float32) * 0.3
+            attn_bias = attn_bias_from_alibi_slopes(alibi_slopes, seqlen, seqlen, causal=causal).to(dtype)
+            attn_bias = repeat(attn_bias, "1 ... -> b ...", b=batch_size)
+            f, b = time_fwd_bwd(
+                flash_attn_func,
+                q, k, v,
+                dropout_p,
+                causal=causal,
+                # alibi_slopes=alibi_slopes,
+                alibi_slopes=None,
+                repeats=repeats,
+                verbose=False
+            )
+            time_f[config, "fa2_baseline"] = f
+            time_b[config, "fa2_baseline"] = b
+
+            q = q.detach().requires_grad_(True)
+            k = k.detach().requires_grad_(True)
+            v = v.detach().requires_grad_(True)
+            f, b = time_fwd_bwd(
+                flash_attn_func,
+                q, k, v,
+                dropout_p,
+                causal=causal,
+                alibi_slopes=rearrange(alibi_slopes, "1 h -> h"),
+                # alibi_slopes=None,
+                repeats=repeats,
+                verbose=False
+            )
+            time_f[config, "fa2_alibi"] = f
+            time_b[config, "fa2_alibi"] = b
+
+            try:
+                q = q.detach().requires_grad_(True)
+                k = k.detach().requires_grad_(True)
+                v = v.detach().requires_grad_(True)
+                f, b = time_fwd_bwd(
+                    attention_pytorch,
+                    q, k, v,
+                    dropout_p,
+                    causal=causal,
+                    attn_bias=attn_bias,
+                    repeats=repeats,
+                    verbose=False
+                )
+            except:  # Skip if OOM
+                f, b = float('nan'), float('nan')
+            time_f[config, "torch"] = f
+            time_b[config, "torch"] = b
+
+            # F.sdpa doesn't currently (torch 2.1) dispatch to flash-attn but just to be safe
+            with torch.backends.cuda.sdp_kernel(enable_flash=False):
+                q_pt = q.detach().requires_grad_(True).transpose(1, 2)
+                k_pt = k.detach().requires_grad_(True).transpose(1, 2)
+                v_pt = v.detach().requires_grad_(True).transpose(1, 2)
+                f, b = time_fwd_bwd(
+                    F.scaled_dot_product_attention,
+                    q_pt, k_pt, v_pt,
+                    attn_mask=attn_bias,
+                    dropout_p=dropout_p,
+                    is_causal=causal,
+                    repeats=repeats,
+                    verbose=False
+                )
+                time_f[config, "sdpa"] = f
+                time_b[config, "sdpa"] = b
+
+            if xops is not None:
+                q = q.detach().requires_grad_(True)
+                k = k.detach().requires_grad_(True)
+                v = v.detach().requires_grad_(True)
+                if causal:
+                    attn_bias_xops = xops.LowerTriangularMask().add_bias(attn_bias.expand(-1, -1, seqlen, -1).to(dtype=q.dtype))
+                    # NotImplementedError: No operator found for `memory_efficient_attention_backward` with inputs:
+                    # `flshattB@v2.3.6` is not supported because:
+                    #     attn_bias type is <class 'xformers.ops.fmha.attn_bias.LowerTriangularMaskWithTensorBias'>
+                    # `cutlassB` is not supported because:
+                    #     attn_bias type is <class 'xformers.ops.fmha.attn_bias.LowerTriangularMaskWithTensorBias'>
+                    attn_bias_xops = attn_bias_xops.materialize((batch_size, nheads, seqlen, seqlen), dtype=q.dtype, device=device)
+                else:
+                    attn_bias_xops = attn_bias.to(dtype=q.dtype)
+                f, b = time_fwd_bwd(
+                    xops.memory_efficient_attention,
+                    q, k, v,
+                    attn_bias_xops,
+                    dropout_p,
+                    repeats=repeats,
+                    verbose=False
+                )
+                time_f[config, "xformers"] = f
+                time_b[config, "xformers"] = b
+
+            q = q.detach().requires_grad_(True)
+            k = k.detach().requires_grad_(True)
+            v = v.detach().requires_grad_(True)
+            cos, sin = generate_cos_sin(seqlen, headdim, device, dtype)
+            f, b = time_fwd_bwd(
+                flash_rotary,
+                q, k, v,
+                cos, sin,
+                causal,
+                repeats=repeats,
+                verbose=False
+            )
+            time_f[config, "fa2_rotary"] = f
+            time_b[config, "fa2_rotary"] = b
+
+            print(f"### causal={causal}, headdim={headdim}, batch_size={batch_size}, seqlen={seqlen} ###")
+            csv_output = ""
+            csv_output += f"{causal},{headdim},{batch_size},{seqlen},"
+            for method in methods:
+                time_f_b[config, method] = time_f[config, method] + time_b[config, method]
+                speed_f[config, method] = efficiency(
+                    flops(batch_size, seqlen, headdim, nheads, causal, mode="fwd"),
+                    time_f[config, method]
+                )
+                speed_b[config, method] = efficiency(
+                    flops(batch_size, seqlen, headdim, nheads, causal, mode="bwd"),
+                    time_b[config, method]
+                )
+                speed_f_b[config, method] = efficiency(
+                    flops(batch_size, seqlen, headdim, nheads, causal, mode="fwd_bwd"),
+                    time_f_b[config, method]
+                )
+                print(
+                    f"{method} fwd: {speed_f[config, method]:.2f} TFLOPs/s, "
+                    f"bwd: {speed_b[config, method]:.2f} TFLOPs/s, "
+                    f"fwd + bwd: {speed_f_b[config, method]:.2f} TFLOPs/s"
+                )
+                csv_output += f"{speed_f[config, method]:.2f},{speed_b[config, method]:.2f},{speed_f_b[config, method]:.2f},"
+            print(csv_output)

Code/Baselines/flash-attention/benchmarks/benchmark_attn.py

@@ -0,0 +1,402 @@
+from collections import namedtuple
+from functools import partial
+import math
+import os
+from typing import NamedTuple
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+import time
+
+try:
+    import cudnn
+except ImportError:
+    cudnn = None
+# cudnn = None
+
+Timing = NamedTuple('timing', [('mean', float)])
+
+
+from einops import rearrange, repeat
+
+# from flash_attn.utils.benchmark import benchmark_forward, benchmark_backward, benchmark_combined, benchmark_all, benchmark_fwd_bwd, pytorch_profiler
+from flash_attn.utils.benchmark import benchmark_forward, benchmark_backward, benchmark_combined, benchmark_all, benchmark_fwd_bwd, pytorch_profiler
+from flash_attn.flash_attn_interface import flash_attn_func, flash_attn_varlen_func
+from flash_attn.cute.interface import flash_attn_func as flash_attn_func_python
+from flash_attn.cute.interface import flash_attn_varlen_func as flash_attn_varlen_func_python
+try:
+    from flash_attn_interface import flash_attn_func as flash_attn_func_v3
+    from flash_attn_interface import flash_attn_varlen_func as flash_attn_varlen_func_v3
+except ImportError:
+    flash_attn_func_v3 = None
+    flash_attn_varlen_func_v3 = None
+
+if torch.cuda.get_device_capability()[0] != 9:
+    flash_attn_func_v3 = None
+# flash_attn_func_v3 = None
+
+flash_attn_func = None
+
+from triton.testing import do_bench
+
+def time_fwd(func, *args, repeats=30, verbose=True, desc="", **kwargs):
+    # # Warmup
+    # for _ in range(5):
+    #     func(*args, **kwargs)
+    # time.sleep(1)
+    # return benchmark_forward(func, *args, **kwargs, repeats=repeats, verbose=verbose, desc=desc)[1]
+    # s = torch.cuda.Stream()
+    # s.wait_stream(torch.cuda.current_stream())
+    # with torch.cuda.stream(s):
+    #     for _ in range(2):
+    #         out = func(*args, **kwargs)
+    # torch.cuda.current_stream().wait_stream(s)
+    # graph = torch.cuda.CUDAGraph()
+    # with torch.cuda.graph(graph):
+    #     out = func(*args, **kwargs)
+    # time_f = benchmark_forward(lambda: graph.replay(), repeats=repeats, verbose=verbose, desc=desc)
+    # # return time_f[1].mean
+    # return time_f[1]
+    return Timing(do_bench(lambda: func(*args, **kwargs), warmup=5, rep=repeats) * 1e-3)
+
+
+def flops(batch, nheads, seqlen_q, seqlen_k, headdim, headdim_v, causal=False, window_size=(None, None)):
+    if causal:
+        avg_seqlen = (max(0, seqlen_k - seqlen_q) + seqlen_k) / 2
+    else:
+        if window_size == (None, None):
+            avg_seqlen = seqlen_k
+        else:
+            row_idx = torch.arange(seqlen_q, device='cuda')
+            col_left = torch.maximum(row_idx + seqlen_k - seqlen_q - window_size[0], torch.tensor(0)) if window_size[0] is not None else torch.zeros_like(row_idx)
+            col_right = torch.minimum(row_idx + seqlen_k - seqlen_q - window_size[1], torch.tensor(seqlen_k - 1)) if window_size[1] is not None else torch.full_like(row_idx, seqlen_k - 1)
+            avg_seqlen = (col_right - col_left + 1).float().mean().item()
+    return batch * nheads * 2 * seqlen_q * avg_seqlen * (headdim + headdim_v)
+
+
+def convert_to_cudnn_type(torch_type):
+    if torch_type == torch.float16:
+        return cudnn.data_type.HALF
+    elif torch_type == torch.bfloat16:
+        return cudnn.data_type.BFLOAT16
+    elif torch_type == torch.float32:
+        return cudnn.data_type.FLOAT
+    elif torch_type == torch.int32:
+        return cudnn.data_type.INT32
+    elif torch_type == torch.int64:
+        return cudnn.data_type.INT64
+    else:
+        raise ValueError("Unsupported tensor data type.")
+
+
+def cudnn_spda_setup(q, k, v, causal=False, window_size_left=None):
+    b, nheads, seqlen_q, headdim = q.shape
+    _, nheads_k, seqlen_k, _ = k.shape
+    assert v.shape == (b, nheads_k, seqlen_k, headdim)
+    assert cudnn is not None, 'CUDNN is not available'
+    q_gpu, k_gpu, v_gpu = q, k, v
+    o_gpu = torch.empty_like(q_gpu)
+    stats_gpu = torch.empty(b, nheads, seqlen_q, 1, dtype=torch.float32, device=q.device)
+    graph = cudnn.pygraph(
+        io_data_type=convert_to_cudnn_type(q.dtype),
+        intermediate_data_type=cudnn.data_type.FLOAT,
+        compute_data_type=cudnn.data_type.FLOAT,
+    )
+    q = graph.tensor_like(q_gpu.detach())
+    k = graph.tensor_like(k_gpu.detach())
+    v = graph.tensor_like(v_gpu.detach())
+
+    o, stats = graph.sdpa(
+        name="sdpa",
+        q=q,
+        k=k,
+        v=v,
+        is_inference=False,
+        attn_scale=1.0 / math.sqrt(headdim),
+        # use_causal_mask_bottom_right=causal or window_size_left is not None,
+        use_causal_mask=causal or window_size_left is not None,
+        sliding_window_length=window_size_left if window_size_left is not None and not causal else None,
+    )
+
+    o.set_output(True).set_dim(o_gpu.shape).set_stride(o_gpu.stride())
+    stats.set_output(True).set_data_type(cudnn.data_type.FLOAT)
+
+    graph.validate()
+    graph.build_operation_graph()
+    graph.create_execution_plans([cudnn.heur_mode.A, cudnn.heur_mode.FALLBACK])
+    graph.check_support()
+    graph.build_plans()
+
+    variant_pack = {
+        q: q_gpu,
+        k: k_gpu,
+        v: v_gpu,
+        o: o_gpu,
+        stats: stats_gpu,
+    }
+
+    workspace = torch.empty(graph.get_workspace_size(), device="cuda", dtype=torch.uint8)
+
+    def run(*args, **kwargs):
+        graph.execute(variant_pack, workspace)
+        return o_gpu
+
+    return run
+
+
+def cudnn_spda_bwd_setup(q, k, v, o, g, lse, causal=False, window_size_left=None):
+    b, nheads, seqlen_q, headdim = q.shape
+    _, nheads_k, seqlen_k, _ = k.shape
+    assert v.shape == (b, nheads_k, seqlen_k, headdim)
+    assert g.shape == (b, nheads, seqlen_q, headdim)
+    assert o.shape == (b, nheads, seqlen_q, headdim)
+    assert lse.shape == (b, nheads, seqlen_q, 1)
+    assert cudnn is not None, 'CUDNN is not available'
+    q_gpu, k_gpu, v_gpu, o_gpu, g_gpu = q, k, v, o, g
+    dq_gpu = torch.empty_like(q_gpu)
+    dk_gpu = torch.empty_like(k_gpu)
+    dv_gpu = torch.empty_like(v_gpu)
+    graph = cudnn.pygraph(
+        io_data_type=convert_to_cudnn_type(q.dtype),
+        intermediate_data_type=cudnn.data_type.FLOAT,
+        compute_data_type=cudnn.data_type.FLOAT,
+    )
+    q = graph.tensor_like(q_gpu.detach())
+    k = graph.tensor_like(k_gpu.detach())
+    v = graph.tensor_like(v_gpu.detach())
+    o = graph.tensor_like(o_gpu.detach())
+    g = graph.tensor_like(g_gpu.detach())
+    stats = graph.tensor_like(lse.detach())
+
+    dq, dk, dv = graph.sdpa_backward(
+        name="sdpa_backward",
+        q=q,
+        k=k,
+        v=v,
+        o=o,
+        dO=g,
+        stats=stats,
+        attn_scale=1.0 / math.sqrt(headdim),
+        # use_causal_mask_bottom_right=causal or window_size_left is not None,
+        use_causal_mask=causal or window_size_left is not None,
+        sliding_window_length=window_size_left if window_size_left is not None and not causal else None,
+    )
+
+    dq.set_output(True).set_dim(dq_gpu.shape).set_stride(dq_gpu.stride())
+    dk.set_output(True).set_dim(dk_gpu.shape).set_stride(dk_gpu.stride())
+    dv.set_output(True).set_dim(dv_gpu.shape).set_stride(dv_gpu.stride())
+
+    graph.validate()
+    graph.build_operation_graph()
+    graph.create_execution_plans([cudnn.heur_mode.A, cudnn.heur_mode.FALLBACK])
+    graph.check_support()
+    graph.build_plans()
+
+    variant_pack = {
+        q: q_gpu,
+        k: k_gpu,
+        v: v_gpu,
+        o: o_gpu,
+        g: g_gpu,
+        stats: lse,
+        dq: dq_gpu,
+        dk: dk_gpu,
+        dv: dv_gpu,
+    }
+
+    workspace = torch.empty(graph.get_workspace_size(), device="cuda", dtype=torch.uint8)
+
+    def run(*args, **kwargs):
+        graph.execute(variant_pack, workspace)
+        return dq_gpu, dk_gpu, dv_gpu
+
+    return run
+
+
+torch.manual_seed(0)
+repeats = 10
+dropout_p = 0.0
+causal = False
+dtype = torch.bfloat16
+# dtype = torch.float8_e4m3fn
+dtype_gen = torch.bfloat16 if dtype == torch.float8_e4m3fn else dtype
+device = 'cuda'
+verbose = True
+varlen = False
+has_backward = False
+page_size = None
+softcap = 0.0
+V_colmajor = False
+deterministic = False
+batch_size = 2
+# seqlen = 2048
+seqlen = 8192
+# seqlen = 4096
+# seqlen = 2047
+dim = 2048
+# headdim = 128
+# headdim = 64
+headdim = 256
+# for headdim in [64, 128, 256]:
+# bs_seqlen_vals = [(32, 512), (16, 1024), (8, 2048), (4, 4096), (2, 8192), (1, 16384)]
+# bs_seqlen_vals = [(16, 1024), (8, 2048), (4, 4096), (2, 8192), (1, 16384)]
+# bs_seqlen_vals = [(32, 512), (16, 1024)]
+# bs_seqlen_vals = [(2, 64 * 132)]
+bs_seqlen_vals = [(4, 8192)]
+# bs_seqlen_vals = [(1, 16 * 1024)]
+time_f = {}
+time_b = {}
+
+# for headdim in [64, 128, 256]:
+# for headdim in [64, 96, 128, 192]:
+# for headdim in [64, 96, 128, 192, 256]:
+# for headdim in [64, 96, 128]:
+# for headdim in [64, 128, 256]:
+# for headdim in [64, 96, 128, 192, 256]:
+for headdim in [128]:
+    nheads = dim // headdim
+    # nheads = 128
+    # headdim = 64
+    # batch_size = 64
+    # seqlen = 512
+    # nheads = 8
+    # headdim = 128
+    nheads_kv = nheads
+    # nheads_kv = nheads // 4
+    # nheads_kv = 1
+    headdim_v = headdim
+    # headdim_v = 512
+    has_qv = headdim == 64 and headdim_v == 512
+    # has_qv = False
+
+    for batch_size, seqlen in bs_seqlen_vals:
+        num_splits = 0
+        # window_size = (-1, -1)
+        window_size = (None, None)
+        window_size_fa = (-1, -1)
+        # window_size = (seqlen // 2 - 1, 0)
+        pack_gqa = None
+        # seqlen_q = 64
+        seqlen_q = seqlen
+        leftpad_k = None
+        # leftpad_k = torch.full((batch_size,), 0, device=device, dtype=torch.int32)
+        q = torch.randn(batch_size, seqlen_q, nheads, headdim, device=device, dtype=dtype_gen, requires_grad=has_backward)
+        k = torch.randn(batch_size, seqlen, nheads_kv, headdim, device=device, dtype=dtype_gen, requires_grad=has_backward)
+        v = torch.randn(batch_size, seqlen, nheads_kv, headdim_v, device=device, dtype=dtype_gen, requires_grad=has_backward)
+        q, k, v = [x.detach().to(dtype).requires_grad_(has_backward) for x in [q, k, v]]
+        v_colmajor = v.detach().transpose(-1, -3).contiguous().transpose(-1, -3).requires_grad_(has_backward)
+        v_fa3 = v if not V_colmajor else v_colmajor
+        qv = torch.randn(batch_size, seqlen_q, nheads, headdim_v, device=device, dtype=dtype_gen) if has_qv else None
+        # q = torch.randint(-2, 3, (batch_size, seqlen, nheads, headdim), device=device, dtype=torch.int32).to(dtype)
+        # k = torch.randint(-2, 3, (batch_size, seqlen, nheads, headdim), device=device, dtype=torch.int32).to(dtype)
+        # v = torch.randint(-2, 3, (batch_size, seqlen, nheads, headdim_v), device=device, dtype=torch.int32).to(dtype)
+        g = torch.randn(batch_size, seqlen_q, nheads, headdim_v, device=device, dtype=dtype_gen)
+        o = torch.randn(batch_size, seqlen_q, nheads, headdim_v, device=device, dtype=dtype_gen)
+        stats = torch.randn(batch_size, seqlen_q, nheads, 1, device=device, dtype=torch.float32)
+        if varlen:
+            q_unpad, k_unpad, v_unpad = [rearrange(x.detach(), "b s h d -> (b s) h d").requires_grad_(has_backward) for x in [q, k, v]]
+            cu_seqlens_q = torch.arange(batch_size + 1, device=device, dtype=torch.int32) * seqlen_q
+            cu_seqlens_k = torch.arange(batch_size + 1, device=device, dtype=torch.int32) * seqlen
+            # cu_seqlens_q = torch.tensor([0, 248, 249, 250, 251, 252, 253, 254, 255, 256], device=device, dtype=torch.int32)
+            # q_unpad = q_unpad[:256]
+            # seqlen_q = 256
+            # cu_seqlens_q = torch.tensor([0, 376, 377, 378, 379, 380, 381, 382, 383, 384], device=device, dtype=torch.int32)
+            # q_unpad = q_unpad[:384]
+            # seqlen_q = 384
+        if page_size is not None:
+            assert seqlen % page_size == 0
+            k_paged, v_paged = [rearrange(x, "b (n p) h d -> (b n) p h d", p=page_size) for x in [k, v]]
+            page_table = rearrange(torch.arange(batch_size * seqlen // page_size, device=device, dtype=torch.int32),
+                                   "(b s) -> b s", s=seqlen // page_size)
+        else:
+            page_table = None
+
+        for causal in [False, True]:
+        # for causal in [False]:
+            print(f"\n### {headdim = }, {causal = }, {seqlen = } ###")
+            nFLOPS = flops(batch_size, nheads, seqlen_q, seqlen, headdim if not has_qv else headdim + headdim_v, headdim_v, causal=causal, window_size=window_size)
+            if cudnn is not None:
+            # if False:
+                if headdim <= 256 and dtype != torch.float8_e4m3fn and headdim == headdim_v:
+                    cudnn_spda = cudnn_spda_setup(q.transpose(1, 2), k.transpose(1, 2), v.transpose(1, 2), causal=causal, window_size_left=window_size[0])
+                    cudnn_spda_bwd = cudnn_spda_bwd_setup(q.transpose(1, 2), k.transpose(1, 2), v.transpose(1, 2), o.transpose(1, 2), g.transpose(1, 2), stats.transpose(1, 2), causal=causal, window_size_left=window_size[0])
+            if dtype != torch.float8_e4m3fn and headdim == headdim_v and flash_attn_func is not None:
+            # if False:
+                if not varlen:
+                    m0 = time_fwd(flash_attn_func, q, k, v, dropout_p, causal=causal, window_size=window_size, softcap=softcap, repeats=repeats, verbose=verbose, desc='Fav2')
+                else:
+                    m0 = time_fwd(flash_attn_varlen_func, q_unpad, k_unpad, v_unpad, cu_seqlens_q, cu_seqlens_k, seqlen_q, seqlen, dropout_p, causal=causal, window_size=window_size, softcap=softcap, repeats=repeats, verbose=verbose, desc='Fav2')
+                time_f[(causal, headdim, batch_size, seqlen), "Flash2"] = m0.mean
+                if has_backward:
+                    time.sleep(1)
+                    if not varlen:
+                        _, m0b = benchmark_backward(flash_attn_func, q, k, v, dropout_p, causal=causal, window_size=window_size, softcap=softcap, deterministic=deterministic,
+                                                    repeats=repeats, verbose=False, desc='Fav2')
+                    else:
+                        _, m0b = benchmark_backward(flash_attn_varlen_func, q_unpad, k_unpad, v_unpad, cu_seqlens_q, cu_seqlens_k, seqlen_q, seqlen, dropout_p, causal=causal, window_size=window_size, softcap=softcap, deterministic=deterministic,
+                                                    repeats=repeats, verbose=False, desc='Fav2')
+                    time_b[(causal, headdim, batch_size, seqlen), "Flash2"] = m0b.mean
+            # pytorch_profiler(flash_attn_func, q, k, v, dropout_p, causal=causal, backward=True)
+
+            if cudnn is not None:
+            # if False:
+                if headdim <= 256 and dtype != torch.float8_e4m3fn and headdim == headdim_v:
+                    time.sleep(1) # Sleep to avoid residual power throttling from the previous benchmark
+                    m2 = time_fwd(cudnn_spda, repeats=repeats, verbose=verbose, desc='CuDNN')
+                    time_f[(causal, headdim, batch_size, seqlen), "cuDNN"] = m2.mean
+                    time.sleep(1)
+                    m2b = time_fwd(cudnn_spda_bwd, repeats=repeats, verbose=verbose, desc='CuDNN')
+                    time_b[(causal, headdim, batch_size, seqlen), "cuDNN"] = m2b.mean
+                # pytorch_profiler(cudnn_spda, backward=False)
+                # pytorch_profiler(cudnn_spda_bwd, backward=False)
+            time.sleep(1)
+            if flash_attn_func_v3 is not None:
+                if not varlen:
+                    # m1 = time_fwd(flash_attn_func_v3, q, k if page_size is None else k_paged, v_fa3 if page_size is None else v_paged, cache_leftpad = leftpad_k, page_table=page_table, causal=causal, window_size=window_size, softcap=softcap, num_splits=num_splits, pack_gqa=pack_gqa, repeats=repeats, verbose=verbose, desc='Fav3')
+                    m1 = time_fwd(flash_attn_func_v3, q, k if page_size is None else k_paged, v_fa3 if page_size is None else v_paged, causal=causal, window_size=window_size_fa, softcap=softcap, num_splits=num_splits, pack_gqa=pack_gqa, repeats=repeats, verbose=verbose, desc='Fav3')
+                    # pytorch_profiler(flash_attn_func_v3, q, k if page_size is None else k_paged, v_fa3 if page_size is None else v_paged, page_table=page_table, causal=causal, window_size=window_size, softcap=softcap, num_splits=num_splits, pack_gqa=pack_gqa)
+                else:
+                    m1 = time_fwd(flash_attn_varlen_func_v3, q_unpad, k_unpad, v_unpad, cu_seqlens_q, cu_seqlens_k, seqlen_q, seqlen, causal=causal, window_size=window_size_fa, softcap=softcap, num_splits=num_splits, pack_gqa=pack_gqa, repeats=repeats, verbose=verbose, desc='Fav3')
+                    # pytorch_profiler(flash_attn_varlen_func_v3, q_unpad, k_unpad, v_unpad, cu_seqlens_q, cu_seqlens_k, seqlen_q, seqlen, causal=causal, window_size=window_size, softcap=softcap, num_splits=num_splits)
+                time_f[(causal, headdim, batch_size, seqlen), "Flash3"] = m1.mean
+            if flash_attn_func_python is not None:
+                if not varlen:
+                    m1_py = time_fwd(flash_attn_func_python, q, k if page_size is None else k_paged, v_fa3 if page_size is None else v_paged, causal=causal, window_size=window_size, softcap=softcap, repeats=repeats, verbose=verbose, desc='Fav3 python')
+                else:
+                    m1_py = time_fwd(flash_attn_varlen_func_python, q_unpad, k_unpad, v_unpad, cu_seqlens_q, cu_seqlens_k, causal=causal, window_size=window_size, softcap=softcap, repeats=repeats, verbose=verbose, desc='Fav3 python')
+            if dtype != torch.float8_e4m3fn and headdim == headdim_v and flash_attn_func_v3 is not None and has_backward:
+                time.sleep(1)
+                if not varlen:
+                    _, m1b = benchmark_backward(flash_attn_func_v3, q, k, v, causal=causal, softcap=softcap, repeats=repeats, verbose=False, desc='Fav3')
+                else:
+                    _, m1b = benchmark_backward(flash_attn_varlen_func_v3, q_unpad, k_unpad, v_unpad, cu_seqlens_q, cu_seqlens_k, seqlen_q, seqlen, causal=causal, window_size=window_size, softcap=softcap, deterministic=deterministic,
+                                                repeats=repeats, verbose=False, desc='Fav3')
+                time_b[(causal, headdim, batch_size, seqlen), "Flash3"] = m1b.mean
+                # time.sleep(1)
+                # if not varlen:
+                #     pytorch_profiler(flash_attn_func_v3, q, k, v, causal=causal, deterministic=deterministic, backward=True)
+                # else:
+                #     pytorch_profiler(flash_attn_varlen_func_v3, q_unpad, k_unpad, v_unpad, cu_seqlens_q, cu_seqlens_k, seqlen_q, seqlen, causal=causal, deterministic=deterministic, backward=True)
+            # benchmark_forward(torch.clone, k, repeats=repeats, verbose=verbose, desc='Memcpy')
+            if dtype != torch.float8_e4m3fn and headdim == headdim_v and flash_attn_func_python is not None and has_backward:
+                _, m1b_py = benchmark_backward(flash_attn_func_python, q, k, v, causal=causal, softcap=softcap, repeats=repeats, verbose=False, desc='Fav2 python')
+
+            if dtype != torch.float8_e4m3fn and headdim == headdim_v and flash_attn_func is not None:
+            # if False:
+                print(f'FAv2 fwd: {m0.mean * 1e3:.3f}ms, {(nFLOPS / m0.mean * 1e-12):.1f} TFLOPS')
+                if has_backward:
+                    print(f'FAv2 bwd: {m0b.mean * 1e3:.3f}ms, {(2.5 * nFLOPS / m0b.mean * 1e-12):.1f} TFLOPS')
+            if cudnn is not None:
+                print(f'CuDNN fwd: {m2.mean * 1e3:.3f}ms, {(nFLOPS / m2.mean * 1e-12):.1f} TFLOPS')
+                if has_backward:
+                    print(f'CuDNN bwd: {m2b.mean * 1e3:.3f}ms, {(2.5 * nFLOPS / m2b.mean * 1e-12):.1f} TFLOPS')
+            if flash_attn_func_v3 is not None:
+                print(f'FAv3 fwd: {m1.mean * 1e3:.3f}ms, {(nFLOPS / m1.mean * 1e-12):.1f} TFLOPS')
+                if dtype != torch.float8_e4m3fn and headdim == headdim_v and has_backward:
+                    print(f'FAv3 bwd: {m1b.mean * 1e3:.3f}ms, {(2.5 * nFLOPS / m1b.mean * 1e-12):.1f} TFLOPS')
+
+            if flash_attn_func_python is not None:
+                print(f'FA Python fwd: {m1_py.mean * 1e3:.3f}ms, {(nFLOPS / m1_py.mean * 1e-12):.1f} TFLOPS')
+                if dtype != torch.float8_e4m3fn and headdim == headdim_v and has_backward:
+                    print(f'FAv2 Python bwd: {m1b_py.mean * 1e3:.3f}ms, {(2.5 * nFLOPS / m1b_py.mean * 1e-12):.1f} TFLOPS')

Code/Baselines/flash-attention/benchmarks/benchmark_causal.py

@@ -0,0 +1,225 @@
+from functools import partial
+import math
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from einops import rearrange, repeat
+
+# from flash_attn.utils.benchmark import benchmark_forward, benchmark_backward, benchmark_combined, benchmark_all, benchmark_fwd_bwd, pytorch_profiler
+from flash_attn.utils.benchmark import benchmark_forward, benchmark_backward, benchmark_combined, benchmark_all, benchmark_fwd_bwd, pytorch_profiler
+from flash_attn.flash_attn_interface import flash_attn_varlen_qkvpacked_func
+# # from flash_attn.triton.fused_attention import attention as attention
+# from flash_attn.flash_attn_triton import flash_attn_qkvpacked_func
+# from flash_attn.flash_attn_triton_og import attention as attention_og
+
+# from triton.ops.flash_attention import attention as attention_triton
+
+from flash_attn import flash_attn_qkvpacked_func, flash_attn_kvpacked_func
+
+try:
+    from flash_attn.fused_softmax import scaled_upper_triang_masked_softmax
+except ImportError:
+    scaled_upper_triang_masked_softmax = None
+
+
+def attention_pytorch(qkv, dropout_p=0.0, causal=True):
+    """
+    Arguments:
+        qkv: (batch_size, seqlen, 3, nheads, head_dim)
+        dropout_p: float
+    Output:
+        output: (batch_size, seqlen, nheads, head_dim)
+    """
+    batch_size, seqlen, _, nheads, d = qkv.shape
+    q, k, v = qkv.unbind(dim=2)
+    q = rearrange(q, 'b t h d -> (b h) t d')
+    k = rearrange(k, 'b s h d -> (b h) d s')
+    softmax_scale = 1.0 / math.sqrt(d)
+    # Preallocate attn_weights for `baddbmm`
+    scores = torch.empty(batch_size * nheads, seqlen, seqlen, dtype=qkv.dtype, device=qkv.device)
+    scores = rearrange(torch.baddbmm(scores, q, k, beta=0, alpha=softmax_scale),
+                       '(b h) t s -> b h t s', h=nheads)
+    if causal:
+        # "triu_tril_cuda_template" not implemented for 'BFloat16'
+        # So we have to construct the mask in float
+        causal_mask = torch.triu(torch.full((seqlen, seqlen), -10000.0, device=scores.device), 1)
+        # TD [2022-09-30]: Adding is faster than masked_fill_ (idk why, just better kernel I guess)
+        scores = scores + causal_mask.to(dtype=scores.dtype)
+    attention = torch.softmax(scores, dim=-1)
+    attention_drop = F.dropout(attention, dropout_p)
+    output = torch.einsum('bhts,bshd->bthd', attention_drop , v)
+    return output.to(dtype=qkv.dtype)
+
+
+def attention_megatron(qkv):
+    """
+    Arguments:
+        qkv: (batch_size, seqlen, 3, nheads, head_dim)
+    Output:
+        output: (batch_size, seqlen, nheads, head_dim)
+    """
+    batch_size, seqlen, _, nheads, d = qkv.shape
+    q, k, v = qkv.unbind(dim=2)
+    q = rearrange(q, 'b t h d -> (b h) t d')
+    k = rearrange(k, 'b s h d -> (b h) d s')
+    softmax_scale = 1.0 / math.sqrt(d)
+    # Preallocate attn_weights for `baddbmm`
+    scores = torch.empty(batch_size * nheads, seqlen, seqlen, dtype=qkv.dtype, device=qkv.device)
+    scores = rearrange(torch.baddbmm(scores, q, k, beta=0, alpha=softmax_scale),
+                       '(b h) t s -> b h t s', h=nheads)
+    attention = scaled_upper_triang_masked_softmax(scores, None, scale=1.0)
+    output = torch.einsum('bhts,bshd->bthd', attention, v)
+    return output.to(dtype=qkv.dtype)
+
+
+torch.manual_seed(0)
+repeats = 30
+batch_size = 8
+seqlen = 2048
+nheads = 12
+headdim = 128
+# nheads = 24
+# headdim = 64
+# batch_size = 64
+# seqlen = 512
+# nheads = 8
+# headdim = 128
+dropout_p = 0.0
+causal = True
+dtype = torch.float16
+device = 'cuda'
+
+qkv = torch.randn(batch_size, seqlen, 3, nheads, headdim, device=device, dtype=dtype,
+                  requires_grad=True)
+cu_seqlens = torch.arange(0, (batch_size + 1) * seqlen, step=seqlen, dtype=torch.int32,
+                          device=qkv.device)
+
+qkv_unpad = rearrange(qkv, 'b s ... -> (b s) ...').detach().requires_grad_(True)
+# benchmark_all(flash_attn_varlen_qkvpacked_func, qkv_unpad,
+#               cu_seqlens, seqlen, dropout_p, causal=causal, repeats=repeats, desc='FlashAttention')
+# pytorch_profiler(flash_attn_varlen_qkvpacked_func, qkv_unpad,
+#                  cu_seqlens, seqlen, dropout_p, causal=causal, backward=True)
+benchmark_forward(flash_attn_qkvpacked_func, qkv, dropout_p, causal=causal, repeats=repeats, desc='Fav2')
+pytorch_profiler(flash_attn_qkvpacked_func, qkv, dropout_p, causal=causal, backward=False)
+
+# for dropout_p in [0.1, 0.0]:
+#     for causal in [False, True]:
+#         print(f"### {dropout_p = }, {causal = } ###")
+#         pytorch_profiler(fav2_qkvpacked_func, qkv, dropout_p, causal=causal, backward=True)
+
+
+# nheads_k = 2
+# q = torch.randn(batch_size, seqlen, nheads, headdim, device=device, dtype=dtype, requires_grad=True)
+# kv = torch.randn(batch_size, seqlen, 2, nheads_k, headdim, device=device, dtype=dtype,
+#                  requires_grad=True)
+# if fav2_kvpacked_func is not None:
+#     benchmark_all(fav2_kvpacked_func, q, kv, dropout_p, causal=causal, repeats=repeats, desc='Fav2')
+#     pytorch_profiler(fav2_kvpacked_func, q, kv, dropout_p, causal=causal, backward=True)
+
+# dropout_p = 0.0
+# causal = False
+# benchmark_all(attention_pytorch, qkv, dropout_p, causal=causal,
+#               repeats=repeats, desc='PyTorch Attention')
+
+# benchmark_all(flash_attn_qkvpacked_func, qkv, None, causal, repeats=repeats, desc='FlashAttention Triton')
+# pytorch_profiler(flash_attn_qkvpacked_func, qkv, None, causal, backward=True)
+
+# q, k, v = [torch.randn(batch_size, nheads, seqlen, headdim, device=device, dtype=dtype,
+#                        requires_grad=True) for _ in range(3)]
+# benchmark_all(attention_og, q, k, v, 1.0, repeats=repeats, desc='FlashAttention Triton OG')
+# # pytorch_profiler(attention, q, k, v, 1.0, backward=True)
+
+# if scaled_upper_triang_masked_softmax is not None:
+#     benchmark_all(attention_megatron, qkv, repeats=repeats, desc='Megatron Attention')
+
+# from src.ops.fftconv import fftconv_func
+
+# dim = nheads * headdim
+# u = torch.randn(batch_size, dim, seqlen, device=device, dtype=dtype, requires_grad=True)
+# k = torch.randn(dim, seqlen, device=device, requires_grad=True)
+# D = torch.randn(dim, device=device, requires_grad=True)
+# benchmark_all(fftconv_func, u, k, D, repeats=repeats, desc='FFTConv')
+# pytorch_profiler(fftconv_func, u, k, D, backward=True)
+# pytorch_profiler(torch.fft.rfft, u.float())
+
+flops = 4 * batch_size * seqlen ** 2 * nheads * headdim
+ideal_a100_time = flops / 312 / 1e9
+print(f"Ideal A100 fwd time: {ideal_a100_time:.3f}ms, bwd time: {ideal_a100_time * 2.5:.3f}ms")
+exit(0)
+
+
+def time_fwd_bwd(func, *args, **kwargs):
+    time_f, time_b = benchmark_fwd_bwd(func, *args, **kwargs)
+    return time_f[1].mean, time_b[1].mean
+
+bs_seqlen_vals = [(32, 512), (16, 1024), (8, 2048), (4, 4096), (2, 8192), (1, 16384)]
+causal_vals = [False, True]
+headdim_vals = [64, 128]
+dim = 2048
+dropout_p = 0.0
+
+time_f = {}
+time_b = {}
+for causal in causal_vals:
+    for headdim in headdim_vals:
+        for batch_size, seqlen in bs_seqlen_vals:
+            nheads = dim // headdim
+            qkv = torch.randn(batch_size, seqlen, 3, nheads, headdim, device=device, dtype=dtype,
+                              requires_grad=True)
+            cu_seqlens = torch.arange(0, (batch_size + 1) * seqlen, step=seqlen, dtype=torch.int32,
+                                    device=qkv.device)
+            qkv_unpad = rearrange(qkv, 'b s ... -> (b s) ...').detach().requires_grad_(True)
+            f, b = time_fwd_bwd(
+                flash_attn_varlen_qkvpacked_func, qkv_unpad, cu_seqlens, seqlen, dropout_p,
+                causal=causal, repeats=repeats, verbose=False
+            )
+            time_f[(causal, headdim, batch_size, seqlen), "Flash"] = f
+            time_b[(causal, headdim, batch_size, seqlen), "Flash"] = b
+
+            qkv = qkv.detach().requires_grad_(True)
+            f, b = time_fwd_bwd(
+                fav2_qkvpacked_func, qkv, dropout_p, causal=causal, repeats=repeats, verbose=False
+            )
+            time_f[(causal, headdim, batch_size, seqlen), "Flash2"] = f
+            time_b[(causal, headdim, batch_size, seqlen), "Flash2"] = b
+
+            # q, k, v = [torch.randn(batch_size, nheads, seqlen, headdim, device=device, dtype=dtype,
+            #                        requires_grad=True) for _ in range(3)]
+            # # Try both values of sequence_parallel and pick the faster one
+            # f, b = time_fwd_bwd(
+            #     attention_triton, q, k, v, causal, headdim**(-0.5),
+            #     False, repeats=repeats, verbose=False
+            # )
+            # _, b0 = time_fwd_bwd(
+            #     attention_triton, q, k, v, causal, headdim**(-0.5),
+            #     True, repeats=repeats, verbose=False
+            # )
+            # time_f[(causal, headdim, batch_size, seqlen), "Triton"] = f
+            # time_b[(causal, headdim, batch_size, seqlen), "Triton"] = min(b, b0)
+
+            if seqlen <= 8 * 1024:
+                qkv = qkv.detach().requires_grad_(True)
+                f, b = time_fwd_bwd(
+                    attention_pytorch, qkv, dropout_p, causal=causal, repeats=repeats, verbose=False
+                )
+            else:
+                f, b = float('nan'), float('nan')
+            time_f[(causal, headdim, batch_size, seqlen), "Pytorch"] = f
+            time_b[(causal, headdim, batch_size, seqlen), "Pytorch"] = b
+
+            # q, k, v = [torch.randn(batch_size, seqlen, nheads, headdim, device=device, dtype=dtype,
+            #                        requires_grad=True) for _ in range(3)]
+            # import xformers.ops as xops
+            # f, b = time_fwd_bwd(
+            #     xops.memory_efficient_attention, q, k, v,
+            #     attn_bias=xops.LowerTriangularMask() if causal else None,
+            #     op=(xops.fmha.cutlass.FwOp, xops.fmha.cutlass.BwOp)
+            # )
+            # time_f[(causal, headdim, batch_size, seqlen), "xformers"] = f
+            # time_b[(causal, headdim, batch_size, seqlen), "xformers"] = b
+
+
+import pickle
+with open('flash2_attn_time_h100.plk', 'wb') as fp:
+    pickle.dump((time_f, time_b), fp, protocol=pickle.HIGHEST_PROTOCOL)

Code/Baselines/flash-attention/benchmarks/benchmark_flash_attention.py

@@ -0,0 +1,180 @@
+# Install the newest triton version with
+# pip install "git+https://github.com/openai/triton.git#egg=triton&subdirectory=python"
+import pickle
+import math
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from einops import rearrange, repeat
+
+from flash_attn.utils.benchmark import benchmark_all, benchmark_forward, benchmark_backward
+from flash_attn.utils.benchmark import benchmark_fwd_bwd, benchmark_combined
+
+from flash_attn import flash_attn_qkvpacked_func
+
+try:
+    from triton.ops.flash_attention import attention as attention_triton
+except ImportError:
+    attention_triton = None
+
+try:
+    import xformers.ops as xops
+except ImportError:
+    xops = None
+
+
+def flops(batch, seqlen, headdim, nheads, causal, mode="fwd"):
+    assert mode in ["fwd", "bwd", "fwd_bwd"]
+    f = 4 * batch * seqlen**2 * nheads * headdim // (2 if causal else 1)
+    return f if mode == "fwd" else (2.5 * f if mode == "bwd" else 3.5 * f)
+
+def efficiency(flop, time):
+    return (flop / time / 10**12) if not math.isnan(time) else 0.0
+
+
+def attention_pytorch(qkv, dropout_p=0.0, causal=True):
+    """
+    Arguments:
+        qkv: (batch_size, seqlen, 3, nheads, head_dim)
+        dropout_p: float
+    Output:
+        output: (batch_size, seqlen, nheads, head_dim)
+    """
+    batch_size, seqlen, _, nheads, d = qkv.shape
+    q, k, v = qkv.unbind(dim=2)
+    q = rearrange(q, 'b t h d -> (b h) t d')
+    k = rearrange(k, 'b s h d -> (b h) d s')
+    softmax_scale = 1.0 / math.sqrt(d)
+    # Preallocate attn_weights for `baddbmm`
+    scores = torch.empty(batch_size * nheads, seqlen, seqlen, dtype=qkv.dtype, device=qkv.device)
+    scores = rearrange(torch.baddbmm(scores, q, k, beta=0, alpha=softmax_scale),
+                       '(b h) t s -> b h t s', h=nheads)
+    if causal:
+        # "triu_tril_cuda_template" not implemented for 'BFloat16'
+        # So we have to construct the mask in float
+        causal_mask = torch.triu(torch.full((seqlen, seqlen), -10000.0, device=scores.device), 1)
+        # TD [2022-09-30]: Adding is faster than masked_fill_ (idk why, just better kernel I guess)
+        scores = scores + causal_mask.to(dtype=scores.dtype)
+    attention = torch.softmax(scores, dim=-1)
+    attention_drop = F.dropout(attention, dropout_p)
+    output = torch.einsum('bhts,bshd->bthd', attention_drop , v)
+    return output.to(dtype=qkv.dtype)
+
+
+def time_fwd_bwd(func, *args, **kwargs):
+    time_f, time_b = benchmark_fwd_bwd(func, *args, **kwargs)
+    return time_f[1].mean, time_b[1].mean
+
+
+repeats = 30
+device = 'cuda'
+dtype = torch.float16
+
+bs_seqlen_vals = [(32, 512), (16, 1024), (8, 2048), (4, 4096), (2, 8192), (1, 16384)]
+causal_vals = [False, True]
+headdim_vals = [64, 128]
+dim = 2048
+dropout_p = 0.0
+
+methods = (["Flash2", "Pytorch"]
+           + (["Triton"] if attention_triton is not None else [])
+           + (["xformers.c"] if xops is not None else [])
+           + (["xformers.f"] if xops is not None else []))
+
+time_f = {}
+time_b = {}
+time_f_b = {}
+speed_f = {}
+speed_b = {}
+speed_f_b = {}
+for causal in causal_vals:
+    for headdim in headdim_vals:
+        for batch_size, seqlen in bs_seqlen_vals:
+            config = (causal, headdim, batch_size, seqlen)
+            nheads = dim // headdim
+            qkv = torch.randn(batch_size, seqlen, 3, nheads, headdim, device=device, dtype=dtype,
+                              requires_grad=True)
+            f, b = time_fwd_bwd(
+                flash_attn_qkvpacked_func, qkv, dropout_p, causal=causal, repeats=repeats, verbose=False
+            )
+            time_f[config, "Flash2"] = f
+            time_b[config, "Flash2"] = b
+
+            try:
+                qkv = qkv.detach().requires_grad_(True)
+                f, b = time_fwd_bwd(
+                    attention_pytorch, qkv, dropout_p, causal=causal, repeats=repeats, verbose=False
+                )
+            except:  # Skip if OOM
+                f, b = float('nan'), float('nan')
+            time_f[config, "Pytorch"] = f
+            time_b[config, "Pytorch"] = b
+
+            if attention_triton is not None:
+                q, k, v = [torch.randn(batch_size, nheads, seqlen, headdim, device=device, dtype=dtype,
+                                    requires_grad=True) for _ in range(3)]
+                # Try both values of sequence_parallel and pick the faster one
+                try:
+                    f, b = time_fwd_bwd(
+                        attention_triton, q, k, v, causal, headdim**(-0.5),
+                        False, repeats=repeats, verbose=False
+                    )
+                except:
+                    f, b = float('nan'), float('inf')
+                try:
+                    _, b0 = time_fwd_bwd(
+                        attention_triton, q, k, v, causal, headdim**(-0.5),
+                        True, repeats=repeats, verbose=False
+                    )
+                except:
+                    b0 = float('inf')
+                time_f[config, "Triton"] = f
+                time_b[config, "Triton"] = min(b, b0) if min(b, b0) < float('inf') else float('nan')
+
+            if xops is not None:
+                q, k, v = [torch.randn(batch_size, seqlen, nheads, headdim, device=device, dtype=dtype,
+                                    requires_grad=True) for _ in range(3)]
+                f, b = time_fwd_bwd(
+                    xops.memory_efficient_attention, q, k, v,
+                    attn_bias=xops.LowerTriangularMask() if causal else None,
+                    op=(xops.fmha.cutlass.FwOp, xops.fmha.cutlass.BwOp)
+                )
+                time_f[config, "xformers.c"] = f
+                time_b[config, "xformers.c"] = b
+
+            if xops is not None:
+                q, k, v = [torch.randn(batch_size, seqlen, nheads, headdim, device=device, dtype=dtype,
+                                    requires_grad=True) for _ in range(3)]
+                f, b = time_fwd_bwd(
+                    xops.memory_efficient_attention, q, k, v,
+                    attn_bias=xops.LowerTriangularMask() if causal else None,
+                    op=(xops.fmha.flash.FwOp, xops.fmha.flash.BwOp)
+                )
+                time_f[config, "xformers.f"] = f
+                time_b[config, "xformers.f"] = b
+
+            print(f"### causal={causal}, headdim={headdim}, batch_size={batch_size}, seqlen={seqlen} ###")
+            for method in methods:
+                time_f_b[config, method] = time_f[config, method] + time_b[config, method]
+                speed_f[config, method] = efficiency(
+                    flops(batch_size, seqlen, headdim, nheads, causal, mode="fwd"),
+                    time_f[config, method]
+                )
+                speed_b[config, method] = efficiency(
+                    flops(batch_size, seqlen, headdim, nheads, causal, mode="bwd"),
+                    time_b[config, method]
+                )
+                speed_f_b[config, method] = efficiency(
+                    flops(batch_size, seqlen, headdim, nheads, causal, mode="fwd_bwd"),
+                    time_f_b[config, method]
+                )
+                print(
+                    f"{method} fwd: {speed_f[config, method]:.2f} TFLOPs/s, "
+                    f"bwd: {speed_b[config, method]:.2f} TFLOPs/s, "
+                    f"fwd + bwd: {speed_f_b[config, method]:.2f} TFLOPs/s"
+                )
+
+
+# with open('flash2_attn_time.plk', 'wb') as fp:
+#     pickle.dump((speed_f, speed_b, speed_f_b), fp, protocol=pickle.HIGHEST_PROTOCOL)

Code/Baselines/flash-attention/benchmarks/benchmark_gemm.py

@@ -0,0 +1,47 @@
+import time
+import torch
+import torch.utils.benchmark as benchmark
+
+from triton.testing import do_bench
+
+if torch.version.cuda:
+    backendBLAS = "cuBLAS"
+elif torch.version.hip:
+    backendBLAS = "hipBLAS"
+
+def benchmark_forward(fn, *inputs, repeats=10, desc='', verbose=True, **kwinputs):
+    """Use Pytorch Benchmark on the forward pass of an arbitrary function."""
+    if verbose:
+        print(desc, '- Forward pass')
+    t = benchmark.Timer(
+            stmt='fn(*inputs, **kwinputs)',
+            globals={'fn': fn, 'inputs': inputs, 'kwinputs': kwinputs},
+            num_threads=torch.get_num_threads(),
+            )
+    m = t.timeit(repeats)
+    if verbose:
+        print(m)
+    return t, m
+
+
+torch.manual_seed(0)
+repeats = 30
+dtype = torch.bfloat16
+device = 'cuda'
+verbose = False
+m, n = 8192, 8192
+
+tflops_matmul = {}
+tflops_matmul1 = {}
+for k in [512, 1024, 1536, 2048, 2560, 3072, 3584, 4096, 4608, 5120, 5632, 6144, 6656, 7168, 7680, 8192]:
+    a = torch.randn(m, k, device=device, dtype=dtype)
+    b = torch.randn(n, k, device=device, dtype=dtype).transpose(-1, -2)
+    nFLOPS_matmul = 2 * m * n * k
+    time.sleep(2)  # to reduce power throttling
+    timing = benchmark_forward(torch.matmul, a, b, desc=backendBLAS, verbose=verbose, repeats=repeats)[1]
+    tflops_matmul[k] = nFLOPS_matmul / timing.mean * 1e-12
+    print(f'[torch.utils.benchmark] {backendBLAS}, {m = }, {n = }, {k = }: {timing.mean * 1e3:.3f}ms, {tflops_matmul[k]:.1f} TFLOPS')
+    time.sleep(2)  # to reduce power throttling
+    ms = do_bench(lambda: torch.matmul(a, b), warmup=10, rep=repeats)
+    tflops_matmul1[k] = nFLOPS_matmul / ms * 1e-9
+    print(f'[triton.test.do_bench]  {backendBLAS}, {m = }, {n = }, {k = }: {ms:.3f}ms, {tflops_matmul1[k]:.1f} TFLOPS')

Code/Baselines/flash-attention/flash_attn.egg-info/PKG-INFO

@@ -0,0 +1,587 @@
+Metadata-Version: 2.4
+Name: flash_attn
+Version: 2.8.1
+Summary: Flash Attention: Fast and Memory-Efficient Exact Attention
+Home-page: https://github.com/Dao-AILab/flash-attention
+Author: Tri Dao
+Author-email: tri@tridao.me
+Classifier: Programming Language :: Python :: 3
+Classifier: License :: OSI Approved :: BSD License
+Classifier: Operating System :: Unix
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+License-File: AUTHORS
+Requires-Dist: torch
+Requires-Dist: einops
+Dynamic: author
+Dynamic: author-email
+Dynamic: classifier
+Dynamic: description
+Dynamic: description-content-type
+Dynamic: home-page
+Dynamic: license-file
+Dynamic: requires-dist
+Dynamic: requires-python
+Dynamic: summary
+
+# FlashAttention
+This repository provides the official implementation of FlashAttention and
+FlashAttention-2 from the
+following papers.
+
+**FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness**  
+Tri Dao, Daniel Y. Fu, Stefano Ermon, Atri Rudra, Christopher Ré  
+Paper: https://arxiv.org/abs/2205.14135  
+IEEE Spectrum [article](https://spectrum.ieee.org/mlperf-rankings-2022) about our submission to the MLPerf 2.0 benchmark using FlashAttention.
+![FlashAttention](assets/flashattn_banner.jpg)
+
+**FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning**  
+Tri Dao
+
+Paper: https://tridao.me/publications/flash2/flash2.pdf
+
+![FlashAttention-2](assets/flashattention_logo.png)
+
+
+## Usage
+
+We've been very happy to see FlashAttention being widely adopted in such a short
+time after its release. This [page](https://github.com/Dao-AILab/flash-attention/blob/main/usage.md)
+contains a partial list of places where FlashAttention is being used.
+
+FlashAttention and FlashAttention-2 are free to use and modify (see LICENSE).
+Please cite and credit FlashAttention if you use it.
+
+
+## FlashAttention-3 beta release
+FlashAttention-3 is optimized for Hopper GPUs (e.g. H100). 
+
+Blogpost: https://tridao.me/blog/2024/flash3/
+
+Paper: https://tridao.me/publications/flash3/flash3.pdf
+
+![FlashAttention-3 speedup on H100 80GB SXM5 with FP16](assets/flash3_fp16_fwd.png)
+
+This is a beta release for testing / benchmarking before we integrate that with
+the rest of the repo.
+
+Currently released:
+- FP16 / BF16 forward and backward, FP8 forward
+
+Requirements: H100 / H800 GPU, CUDA >= 12.3.
+
+We highly recommend CUDA 12.8 for best performance.
+
+To install:
+```sh
+cd hopper
+python setup.py install
+```
+To run the test:
+```sh
+export PYTHONPATH=$PWD
+pytest -q -s test_flash_attn.py
+```
+Once the package is installed, you can import it as follows:
+```python
+import flash_attn_interface
+flash_attn_interface.flash_attn_func()
+```
+
+## Installation and features
+**Requirements:**
+- CUDA toolkit or ROCm toolkit
+- PyTorch 2.2 and above.
+- `packaging` Python package (`pip install packaging`)
+- `ninja` Python package (`pip install ninja`) *
+- Linux. Might work for Windows starting v2.3.2 (we've seen a few positive [reports](https://github.com/Dao-AILab/flash-attention/issues/595)) but Windows compilation still requires more testing. If you have ideas on how to set up prebuilt CUDA wheels for Windows, please reach out via Github issue.
+
+\* Make sure that `ninja` is installed and that it works correctly (e.g. `ninja
+--version` then `echo $?` should return exit code 0). If not (sometimes `ninja
+--version` then `echo $?` returns a nonzero exit code), uninstall then reinstall
+`ninja` (`pip uninstall -y ninja && pip install ninja`). Without `ninja`,
+compiling can take a very long time (2h) since it does not use multiple CPU
+cores. With `ninja` compiling takes 3-5 minutes on a 64-core machine using CUDA toolkit.
+
+**To install:**
+```sh
+pip install flash-attn --no-build-isolation
+```
+Alternatively you can compile from source:
+```sh
+python setup.py install
+```
+
+If your machine has less than 96GB of RAM and lots of CPU cores, `ninja` might
+run too many parallel compilation jobs that could exhaust the amount of RAM. To
+limit the number of parallel compilation jobs, you can set the environment
+variable `MAX_JOBS`:
+```sh
+MAX_JOBS=4 pip install flash-attn --no-build-isolation
+```
+
+**Interface:** `src/flash_attention_interface.py`
+
+### NVIDIA CUDA Support
+**Requirements:**
+- CUDA 12.0 and above.
+
+We recommend the
+[Pytorch](https://catalog.ngc.nvidia.com/orgs/nvidia/containers/pytorch)
+container from Nvidia, which has all the required tools to install FlashAttention.
+
+FlashAttention-2 with CUDA currently supports:
+1. Ampere, Ada, or Hopper GPUs (e.g., A100, RTX 3090, RTX 4090, H100). Support for Turing
+   GPUs (T4, RTX 2080) is coming soon, please use FlashAttention 1.x for Turing
+   GPUs for now.
+2. Datatype fp16 and bf16 (bf16 requires Ampere, Ada, or Hopper GPUs).
+3. All head dimensions up to 256. ~~Head dim > 192 backward requires A100/A800 or H100/H800~~. Head dim 256 backward now works on consumer GPUs (if there's no dropout) as of flash-attn 2.5.5.
+
+### AMD ROCm Support
+ROCm version has two backends. There is [composable_kernel](https://github.com/ROCm/composable_kernel) (ck) which is the default backend and a [Triton](https://github.com/triton-lang/triton) backend. They provide an implementation of FlashAttention-2.
+
+**Requirements:**
+- ROCm 6.0 and above.
+
+We recommend the
+[Pytorch](https://hub.docker.com/r/rocm/pytorch)
+container from ROCm, which has all the required tools to install FlashAttention.
+
+#### Composable Kernel Backend
+FlashAttention-2 ROCm CK backend currently supports:
+1. MI200 or MI300 GPUs.
+2. Datatype fp16 and bf16
+3. Both forward's and backward's head dimensions up to 256.
+
+#### Triton Backend
+The Triton implementation of the [Flash Attention v2](https://tridao.me/publications/flash2/flash2.pdf) is currently a work in progress.
+
+It supports AMD's CDNA (MI200, MI300) and RDNA GPU's using fp16, bf16 and fp32 datatypes.
+
+These features are supported in Fwd and Bwd
+1) Fwd and Bwd with causal masking
+2) Variable sequence lengths
+3) Arbitrary Q and KV sequence lengths
+4) Arbitrary head sizes
+5) Multi and grouped query attention
+6) Dropout
+7) Rotary embeddings
+8) ALiBi
+
+We are working on the following things
+1) Paged Attention 
+2) Sliding Window
+3) FP8
+4) Performance Improvements
+
+##### Getting Started
+To get started with the triton backend for AMD, follow the steps below.
+
+First install the recommended Triton version 
+
+```
+pip install triton==3.2.0
+```
+Then install Flash Attention with the flag `FLASH_ATTENTION_TRITON_AMD_ENABLE` set to `"TRUE"`.
+
+```
+cd flash-attention
+git checkout main_perf
+FLASH_ATTENTION_TRITON_AMD_ENABLE="TRUE" python setup.py install
+```
+
+To test that things are working, you can run our tests. These tests take hours so you don't need to run the full thing.
+```
+FLASH_ATTENTION_TRITON_AMD_ENABLE="TRUE" pytest tests/test_flash_attn_triton_amd.py
+```
+
+You can use autotune for better performance by using this flag `FLASH_ATTENTION_TRITON_AMD_AUTOTUNE="TRUE"`
+```
+FLASH_ATTENTION_TRITON_AMD_ENABLE="TRUE" FLASH_ATTENTION_TRITON_AMD_AUTOTUNE="TRUE" python $PATH_TO_CODE
+```
+
+###### Docker
+You can also use the Dockerfile below which does the above steps on top of the latest rocm/pytorch image.
+```
+FROM rocm/pytorch:latest
+
+WORKDIR /workspace
+
+# install triton
+RUN pip install triton==3.2.0
+
+# install flash attention
+ENV FLASH_ATTENTION_TRITON_AMD_ENABLE="TRUE"
+
+RUN git clone https://github.com/ROCm/flash-attention.git &&\ 
+    cd flash-attention &&\
+    git checkout main_perf &&\
+    python setup.py install
+
+# set working dir
+WORKDIR /workspace/flash-attention
+```
+
+To build the docker file
+```
+docker build -t fa_triton .
+```
+
+To run the docker image
+```
+docker run -it --network=host --user root --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined --ipc=host --shm-size 16G --device=/dev/kfd --device=/dev/dri fa_triton
+```
+
+## How to use FlashAttention
+
+The main functions implement scaled dot product attention (softmax(Q @ K^T *
+softmax_scale) @ V):
+```python
+from flash_attn import flash_attn_qkvpacked_func, flash_attn_func
+```
+
+```python
+flash_attn_qkvpacked_func(qkv, dropout_p=0.0, softmax_scale=None, causal=False,
+                          window_size=(-1, -1), alibi_slopes=None, deterministic=False):
+"""dropout_p should be set to 0.0 during evaluation
+If Q, K, V are already stacked into 1 tensor, this function will be faster than
+calling flash_attn_func on Q, K, V since the backward pass avoids explicit concatenation
+of the gradients of Q, K, V.
+If window_size != (-1, -1), implements sliding window local attention. Query at position i
+will only attend to keys between [i - window_size[0], i + window_size[1]] inclusive.
+Arguments:
+    qkv: (batch_size, seqlen, 3, nheads, headdim)
+    dropout_p: float. Dropout probability.
+    softmax_scale: float. The scaling of QK^T before applying softmax.
+        Default to 1 / sqrt(headdim).
+    causal: bool. Whether to apply causal attention mask (e.g., for auto-regressive modeling).
+    window_size: (left, right). If not (-1, -1), implements sliding window local attention.
+    alibi_slopes: (nheads,) or (batch_size, nheads), fp32. A bias of (-alibi_slope * |i - j|) is added to
+        the attention score of query i and key j.
+    deterministic: bool. Whether to use the deterministic implementation of the backward pass,
+        which is slightly slower and uses more memory. The forward pass is always deterministic.
+Return:
+    out: (batch_size, seqlen, nheads, headdim).
+"""
+```
+
+```python
+flash_attn_func(q, k, v, dropout_p=0.0, softmax_scale=None, causal=False,
+                window_size=(-1, -1), alibi_slopes=None, deterministic=False):
+"""dropout_p should be set to 0.0 during evaluation
+Supports multi-query and grouped-query attention (MQA/GQA) by passing in KV with fewer heads
+than Q. Note that the number of heads in Q must be divisible by the number of heads in KV.
+For example, if Q has 6 heads and K, V have 2 heads, head 0, 1, 2 of Q will attention to head
+0 of K, V, and head 3, 4, 5 of Q will attention to head 1 of K, V.
+If window_size != (-1, -1), implements sliding window local attention. Query at position i
+will only attend to keys between
+[i + seqlen_k - seqlen_q - window_size[0], i + seqlen_k - seqlen_q + window_size[1]] inclusive.
+
+Arguments:
+    q: (batch_size, seqlen, nheads, headdim)
+    k: (batch_size, seqlen, nheads_k, headdim)
+    v: (batch_size, seqlen, nheads_k, headdim)
+    dropout_p: float. Dropout probability.
+    softmax_scale: float. The scaling of QK^T before applying softmax.
+        Default to 1 / sqrt(headdim).
+    causal: bool. Whether to apply causal attention mask (e.g., for auto-regressive modeling).
+    window_size: (left, right). If not (-1, -1), implements sliding window local attention.
+    alibi_slopes: (nheads,) or (batch_size, nheads), fp32. A bias of
+        (-alibi_slope * |i + seqlen_k - seqlen_q - j|)
+        is added to the attention score of query i and key j.
+    deterministic: bool. Whether to use the deterministic implementation of the backward pass,
+        which is slightly slower and uses more memory. The forward pass is always deterministic.
+Return:
+    out: (batch_size, seqlen, nheads, headdim).
+"""
+```
+
+```python
+def flash_attn_with_kvcache(
+    q,
+    k_cache,
+    v_cache,
+    k=None,
+    v=None,
+    rotary_cos=None,
+    rotary_sin=None,
+    cache_seqlens: Optional[Union[(int, torch.Tensor)]] = None,
+    cache_batch_idx: Optional[torch.Tensor] = None,
+    block_table: Optional[torch.Tensor] = None,
+    softmax_scale=None,
+    causal=False,
+    window_size=(-1, -1),  # -1 means infinite context window
+    rotary_interleaved=True,
+    alibi_slopes=None,
+):
+    """
+    If k and v are not None, k_cache and v_cache will be updated *inplace* with the new values from
+    k and v. This is useful for incremental decoding: you can pass in the cached keys/values from
+    the previous step, and update them with the new keys/values from the current step, and do
+    attention with the updated cache, all in 1 kernel.
+
+    If you pass in k / v, you must make sure that the cache is large enough to hold the new values.
+    For example, the KV cache could be pre-allocated with the max sequence length, and you can use
+    cache_seqlens to keep track of the current sequence lengths of each sequence in the batch.
+
+    Also apply rotary embedding if rotary_cos and rotary_sin are passed in. The key @k will be
+    rotated by rotary_cos and rotary_sin at indices cache_seqlens, cache_seqlens + 1, etc.
+    If causal or local (i.e., window_size != (-1, -1)), the query @q will be rotated by rotary_cos
+    and rotary_sin at indices cache_seqlens, cache_seqlens + 1, etc.
+    If not causal and not local, the query @q will be rotated by rotary_cos and rotary_sin at
+    indices cache_seqlens only (i.e. we consider all tokens in @q to be at position cache_seqlens).
+
+    See tests/test_flash_attn.py::test_flash_attn_kvcache for examples of how to use this function.
+
+    Supports multi-query and grouped-query attention (MQA/GQA) by passing in KV with fewer heads
+    than Q. Note that the number of heads in Q must be divisible by the number of heads in KV.
+    For example, if Q has 6 heads and K, V have 2 heads, head 0, 1, 2 of Q will attention to head
+    0 of K, V, and head 3, 4, 5 of Q will attention to head 1 of K, V.
+
+    If causal=True, the causal mask is aligned to the bottom right corner of the attention matrix.
+    For example, if seqlen_q = 2 and seqlen_k = 5, the causal mask (1 = keep, 0 = masked out) is:
+        1 1 1 1 0
+        1 1 1 1 1
+    If seqlen_q = 5 and seqlen_k = 2, the causal mask is:
+        0 0
+        0 0
+        0 0
+        1 0
+        1 1
+    If the row of the mask is all zero, the output will be zero.
+
+    If window_size != (-1, -1), implements sliding window local attention. Query at position i
+    will only attend to keys between
+    [i + seqlen_k - seqlen_q - window_size[0], i + seqlen_k - seqlen_q + window_size[1]] inclusive.
+
+    Note: Does not support backward pass.
+
+    Arguments:
+        q: (batch_size, seqlen, nheads, headdim)
+        k_cache: (batch_size_cache, seqlen_cache, nheads_k, headdim) if there's no block_table,
+            or (num_blocks, page_block_size, nheads_k, headdim) if there's a block_table (i.e. paged KV cache)
+            page_block_size must be a multiple of 256.
+        v_cache: (batch_size_cache, seqlen_cache, nheads_k, headdim) if there's no block_table,
+            or (num_blocks, page_block_size, nheads_k, headdim) if there's a block_table (i.e. paged KV cache)
+        k [optional]: (batch_size, seqlen_new, nheads_k, headdim). If not None, we concatenate
+            k with k_cache, starting at the indices specified by cache_seqlens.
+        v [optional]: (batch_size, seqlen_new, nheads_k, headdim). Similar to k.
+        rotary_cos [optional]: (seqlen_ro, rotary_dim / 2). If not None, we apply rotary embedding
+            to k and q. Only applicable if k and v are passed in. rotary_dim must be divisible by 16.
+        rotary_sin [optional]: (seqlen_ro, rotary_dim / 2). Similar to rotary_cos.
+        cache_seqlens: int, or (batch_size,), dtype torch.int32. The sequence lengths of the
+            KV cache.
+        block_table [optional]: (batch_size, max_num_blocks_per_seq), dtype torch.int32.
+        cache_batch_idx: (batch_size,), dtype torch.int32. The indices used to index into the KV cache.
+            If None, we assume that the batch indices are [0, 1, 2, ..., batch_size - 1].
+            If the indices are not distinct, and k and v are provided, the values updated in the cache
+                 might come from any of the duplicate indices.
+        softmax_scale: float. The scaling of QK^T before applying softmax.
+            Default to 1 / sqrt(headdim).
+        causal: bool. Whether to apply causal attention mask (e.g., for auto-regressive modeling).
+        window_size: (left, right). If not (-1, -1), implements sliding window local attention.
+        rotary_interleaved: bool. Only applicable if rotary_cos and rotary_sin are passed in.
+            If True, rotary embedding will combine dimensions 0 & 1, 2 & 3, etc. If False,
+            rotary embedding will combine dimensions 0 & rotary_dim / 2, 1 & rotary_dim / 2 + 1
+            (i.e. GPT-NeoX style).
+        alibi_slopes: (nheads,) or (batch_size, nheads), fp32. A bias of
+            (-alibi_slope * |i + seqlen_k - seqlen_q - j|)
+            is added to the attention score of query i and key j.
+
+    Return:
+        out: (batch_size, seqlen, nheads, headdim).
+    """
+```
+
+To see how these functions are used in a multi-head attention layer (which
+includes QKV projection, output projection), see the MHA [implementation](https://github.com/Dao-AILab/flash-attention/blob/main/flash_attn/modules/mha.py).
+
+## Changelog
+
+### 2.0: Complete rewrite, 2x faster
+Upgrading from FlashAttention (1.x) to FlashAttention-2
+
+These functions have been renamed:
+- `flash_attn_unpadded_func` -> `flash_attn_varlen_func`
+- `flash_attn_unpadded_qkvpacked_func` -> `flash_attn_varlen_qkvpacked_func`
+- `flash_attn_unpadded_kvpacked_func` -> `flash_attn_varlen_kvpacked_func`
+
+If the inputs have the same sequence lengths in the same batch, it is simpler
+and faster to use these functions:
+```python
+flash_attn_qkvpacked_func(qkv, dropout_p=0.0, softmax_scale=None, causal=False)
+```
+```python
+flash_attn_func(q, k, v, dropout_p=0.0, softmax_scale=None, causal=False)
+```
+### 2.1: Change behavior of causal flag
+
+If seqlen_q != seqlen_k and causal=True, the causal mask is aligned to the
+bottom right corner of the attention matrix, instead of the top-left corner.
+
+For example, if seqlen_q = 2 and seqlen_k = 5, the causal mask (1 = keep, 0 =
+masked out) is:  
+v2.0:  
+    1 0 0 0 0  
+    1 1 0 0 0  
+v2.1:  
+    1 1 1 1 0  
+    1 1 1 1 1  
+
+If seqlen_q = 5 and seqlen_k = 2, the causal mask is:  
+v2.0:  
+    1 0  
+    1 1  
+    1 1  
+    1 1  
+    1 1  
+v2.1:  
+    0 0  
+    0 0  
+    0 0  
+    1 0  
+    1 1  
+If the row of the mask is all zero, the output will be zero.
+
+### 2.2: Optimize for inference
+
+Optimize for inference (iterative decoding) when query has very small sequence
+length (e.g., query sequence length = 1). The bottleneck here is to load KV
+cache as fast as possible, and we split the loading across different thread
+blocks, with a separate kernel to combine results.
+
+See the function `flash_attn_with_kvcache` with more features for inference
+(perform rotary embedding, updating KV cache inplace).
+
+Thanks to the xformers team, and in particular Daniel Haziza, for this
+collaboration.
+
+### 2.3: Local (i.e., sliding window) attention
+
+Implement sliding window attention (i.e., local attention). Thanks to [Mistral
+AI](https://mistral.ai/) and in particular Timothée Lacroix for this
+contribution. Sliding window was used in the [Mistral 7B](https://mistral.ai/news/announcing-mistral-7b/) model.
+
+### 2.4: ALiBi (attention with linear bias), deterministic backward pass.
+
+Implement ALiBi (Press et al., 2021). Thanks to Sanghun Cho from Kakao Brain for this contribution.
+
+Implement deterministic backward pass. Thanks to engineers from [Meituan](www.meituan.com) for this contribution.
+
+### 2.5: Paged KV cache.
+
+Support paged KV cache (i.e., [PagedAttention](https://arxiv.org/abs/2309.06180)).
+Thanks to @beginlner for this contribution.
+
+### 2.6: Softcapping.
+
+Support attention with softcapping, as used in Gemma-2 and Grok models.
+Thanks to @Narsil and @lucidrains for this contribution.
+
+### 2.7: Compatibility with torch compile
+
+Thanks to @ani300 for this contribution.
+
+## Performance
+
+We present expected speedup (combined forward + backward pass) and memory savings from using FlashAttention against PyTorch standard attention, depending on sequence length, on different GPUs (speedup depends on memory bandwidth - we see more speedup on slower GPU memory).
+
+We currently have benchmarks for these GPUs:
+* [A100](#a100)
+* [H100](#h100)
+<!-- * [RTX 3090](#rtx-3090) -->
+<!-- * [T4](#t4) -->
+
+### A100
+
+We display FlashAttention speedup using these parameters:
+* Head dimension 64 or 128, hidden dimension 2048 (i.e. either 32 or 16 heads).
+* Sequence length 512, 1k, 2k, 4k, 8k, 16k.
+* Batch size set to 16k / seqlen.
+
+#### Speedup
+
+![FlashAttention speedup on A100 80GB SXM5 with FP16/BF16](assets/flash2_a100_fwd_bwd_benchmark.png)
+
+#### Memory
+
+![FlashAttention memory](assets/flashattn_memory.jpg)
+
+We show memory savings in this graph (note that memory footprint is the same no matter if you use dropout or masking).
+Memory savings are proportional to sequence length -- since standard attention has memory quadratic in sequence length, whereas FlashAttention has memory linear in sequence length.
+We see 10X memory savings at sequence length 2K, and 20X at 4K.
+As a result, FlashAttention can scale to much longer sequence lengths.
+
+### H100
+
+![FlashAttention speedup on H100 SXM5 with FP16/BF16](assets/flash2_h100_fwd_bwd_benchmark.png)
+
+## Full model code and training script
+
+We have released the full GPT model
+[implementation](https://github.com/Dao-AILab/flash-attention/blob/main/flash_attn/models/gpt.py).
+We also provide optimized implementations of other layers (e.g., MLP, LayerNorm,
+cross-entropy loss, rotary embedding). Overall this speeds up training by 3-5x
+compared to the baseline implementation from Huggingface, reaching up to 225
+TFLOPs/sec per A100, equivalent to 72% model FLOPs utilization (we don't need
+any activation checkpointing).
+
+We also include a training
+[script](https://github.com/Dao-AILab/flash-attention/tree/main/training) to
+train GPT2 on Openwebtext and GPT3 on The Pile.
+
+## Triton implementation of FlashAttention
+
+Phil Tillet (OpenAI) has an experimental implementation of FlashAttention in Triton:
+https://github.com/openai/triton/blob/master/python/tutorials/06-fused-attention.py
+
+As Triton is a higher-level language than CUDA, it might be easier to understand
+and experiment with. The notations in the Triton implementation are also closer
+to what's used in our paper.
+
+We also have an experimental implementation in Triton that support attention
+bias (e.g. ALiBi):
+https://github.com/Dao-AILab/flash-attention/blob/main/flash_attn/flash_attn_triton.py
+
+
+## Tests
+We test that FlashAttention produces the same output and gradient as a reference
+implementation, up to some numerical tolerance. In particular, we check that the
+maximum numerical error of FlashAttention is at most twice the numerical error
+of a baseline implementation in Pytorch (for different head dimensions, input
+dtype, sequence length, causal / non-causal).
+
+To run the tests:
+```sh
+pytest -q -s tests/test_flash_attn.py
+```
+## When you encounter issues
+
+This new release of FlashAttention-2 has been tested on several GPT-style
+models, mostly on A100 GPUs.
+
+If you encounter bugs, please open a GitHub Issue!
+
+## Tests
+To run the tests:
+```sh
+pytest tests/test_flash_attn_ck.py
+```
+
+## Citation
+If you use this codebase, or otherwise found our work valuable, please cite:
+```
+@inproceedings{dao2022flashattention,
+  title={Flash{A}ttention: Fast and Memory-Efficient Exact Attention with {IO}-Awareness},
+  author={Dao, Tri and Fu, Daniel Y. and Ermon, Stefano and Rudra, Atri and R{\'e}, Christopher},
+  booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
+  year={2022}
+}
+@inproceedings{dao2023flashattention2,
+  title={Flash{A}ttention-2: Faster Attention with Better Parallelism and Work Partitioning},
+  author={Dao, Tri},
+  booktitle={International Conference on Learning Representations (ICLR)},
+  year={2024}
+}
+```

Code/Baselines/flash-attention/flash_attn.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

Code/Baselines/flash-attention/flash_attn.egg-info/requires.txt

@@ -0,0 +1,2 @@
+torch
+einops

Code/Baselines/flash-attention/flash_attn.egg-info/top_level.txt

@@ -0,0 +1,3 @@
+flash_attn
+flash_attn_2_cuda
+hopper

Code/Baselines/flash-attention/hopper/benchmark_flash_attention_fp8.py

@@ -0,0 +1,353 @@
+# Install the newest triton version with
+# pip install "git+https://github.com/openai/triton.git#egg=triton&subdirectory=python"
+import pickle
+import math
+import time
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from einops import rearrange, repeat
+
+from flash_attn.utils.benchmark import benchmark_all, benchmark_forward, benchmark_backward
+from flash_attn.utils.benchmark import benchmark_fwd_bwd, benchmark_combined
+
+from flash_attn import flash_attn_qkvpacked_func
+from flash_attn_interface import flash_attn_func, _flash_attn_forward
+
+try:
+    from triton_fused_attention import attention as attention_triton
+except ImportError:
+    attention_triton = None
+
+try:
+    import xformers.ops as xops
+except ImportError:
+    xops = None
+
+try:
+    import cudnn
+except ImportError:
+    cudnn = None
+
+
+def convert_to_cudnn_type(torch_type):
+    if torch_type == torch.float16:
+        return cudnn.data_type.HALF
+    elif torch_type == torch.bfloat16:
+        return cudnn.data_type.BFLOAT16
+    elif torch_type == torch.float32:
+        return cudnn.data_type.FLOAT
+    elif torch_type == torch.int32:
+        return cudnn.data_type.INT32
+    elif torch_type == torch.int64:
+        return cudnn.data_type.INT64
+    elif torch_type == torch.float8_e4m3fn:
+        return cudnn.data_type.FP8_E4M3
+    elif torch_type == torch.float8_e5m2:
+        return cudnn.data_type.FP8_E5M2
+    else:
+        raise ValueError("Unsupported tensor data type.")
+
+def cudnn_spda_setup(qkv, seqlen_q, seqlen_k, causal=False):
+    b, _, _, nheads, headdim = qkv.shape
+    assert cudnn is not None, 'CUDNN is not available'
+    o_gpu = torch.zeros(b, seqlen_q, nheads, headdim, dtype=qkv.dtype, device=qkv.device)
+    o_gpu_transposed = torch.as_strided(
+        o_gpu,
+        [b, nheads, seqlen_q, headdim],
+        [nheads * seqlen_q * headdim, headdim, nheads * headdim, 1],
+    )
+    stats_gpu = torch.empty(b, nheads, seqlen_q, 1, dtype=torch.float32, device=qkv.device)
+    amax_s_gpu = torch.empty(1, 1, 1, 1, dtype=torch.float32, device=qkv.device)
+    amax_o_gpu = torch.empty(1, 1, 1, 1, dtype=torch.float32, device=qkv.device)
+    graph = cudnn.pygraph(
+        io_data_type=convert_to_cudnn_type(qkv.dtype),
+        intermediate_data_type=cudnn.data_type.FLOAT,
+        compute_data_type=cudnn.data_type.FLOAT,
+    )
+    new_q = torch.as_strided(
+        qkv,
+        [b, nheads, seqlen_q, headdim],
+        [seqlen_q * nheads * headdim * 3, headdim, headdim * nheads * 3, 1],
+        storage_offset=0,
+    )
+    q = graph.tensor(
+        name = "Q",
+        dim = list(new_q.shape),
+        stride = list(new_q.stride()),
+        data_type=convert_to_cudnn_type(qkv.dtype)
+    )
+    new_k = torch.as_strided(
+        qkv,
+        [b, nheads, seqlen_k, headdim],
+        [seqlen_k * nheads * headdim * 3, headdim, headdim * nheads * 3, 1],
+        storage_offset=nheads * headdim,
+    )
+    k = graph.tensor(
+        name = "K",
+        dim = list(new_k.shape),
+        stride = list(new_k.stride()),
+        data_type=convert_to_cudnn_type(qkv.dtype)
+    )
+    new_v = torch.as_strided(
+        qkv,
+        [b, nheads, seqlen_k, headdim],
+        [seqlen_k * nheads * headdim * 3, headdim, headdim * nheads * 3, 1],
+        storage_offset=nheads * headdim * 2,
+    )
+    v = graph.tensor(
+        name = "V",
+        dim = list(new_v.shape),
+        stride = list(new_v.stride()),
+        data_type=convert_to_cudnn_type(qkv.dtype)
+    )
+
+    def get_default_scale_tensor():
+        return graph.tensor(
+            dim = [1, 1, 1, 1],
+            stride = [1, 1, 1, 1],
+            data_type=cudnn.data_type.FLOAT
+        )
+
+    default_scale_gpu = torch.ones(1, 1, 1, 1, dtype=torch.float32, device="cuda")
+    descale_q = get_default_scale_tensor()
+    descale_k = get_default_scale_tensor()
+    descale_v = get_default_scale_tensor()
+    descale_s = get_default_scale_tensor()
+    scale_s = get_default_scale_tensor()
+    scale_o = get_default_scale_tensor()
+
+    o, _, amax_s, amax_o = graph.sdpa_fp8(
+        q=q,
+        k=k,
+        v=v,
+        descale_q=descale_q,
+        descale_k=descale_k,
+        descale_v=descale_v,
+        descale_s=descale_s,
+        scale_s=scale_s,
+        scale_o=scale_o,
+        is_inference=True,
+        attn_scale=1.0 / math.sqrt(headdim),
+        use_causal_mask=causal,
+        name="sdpa",
+    )
+
+    o.set_output(True).set_dim(o_gpu_transposed.shape).set_stride(o_gpu_transposed.stride())
+
+    amax_s.set_output(False).set_dim(amax_s_gpu.shape).set_stride(amax_s_gpu.stride())
+    amax_o.set_output(False).set_dim(amax_o_gpu.shape).set_stride(amax_o_gpu.stride())
+    # stats.set_output(True).set_data_type(cudnn.data_type.FLOAT)
+
+    graph.validate()
+    graph.build_operation_graph()
+    graph.create_execution_plans([cudnn.heur_mode.A, cudnn.heur_mode.FALLBACK])
+    graph.check_support()
+    graph.build_plans()
+
+    variant_pack = {
+        q: new_q,
+        k: new_k,
+        v: new_v,
+        descale_q: default_scale_gpu,
+        descale_k: default_scale_gpu,
+        descale_v: default_scale_gpu,
+        descale_s: default_scale_gpu,
+        scale_s: default_scale_gpu,
+        scale_o: default_scale_gpu,
+        o: o_gpu_transposed,
+        amax_s: amax_s_gpu,
+        amax_o: amax_o_gpu,
+    }
+
+    workspace = torch.empty(graph.get_workspace_size(), device="cuda", dtype=torch.uint8)
+
+    def run(*args, **kwargs):
+        graph.execute(variant_pack, workspace)
+        return o_gpu, amax_o_gpu
+
+    return run
+
+
+def attention_pytorch(qkv, dropout_p=0.0, causal=True):
+    """
+    Arguments:
+        qkv: (batch_size, seqlen, 3, nheads, head_dim)
+        dropout_p: float
+    Output:
+        output: (batch_size, seqlen, nheads, head_dim)
+    """
+    batch_size, seqlen, _, nheads, d = qkv.shape
+    q, k, v = qkv.unbind(dim=2)
+    q = rearrange(q, 'b t h d -> (b h) t d')
+    k = rearrange(k, 'b s h d -> (b h) d s')
+    softmax_scale = 1.0 / math.sqrt(d)
+    # Preallocate attn_weights for `baddbmm`
+    scores = torch.empty(batch_size * nheads, seqlen, seqlen, dtype=qkv.dtype, device=qkv.device)
+    scores = rearrange(torch.baddbmm(scores, q, k, beta=0, alpha=softmax_scale),
+                       '(b h) t s -> b h t s', h=nheads)
+    if causal:
+        # "triu_tril_cuda_template" not implemented for 'BFloat16'
+        # So we have to construct the mask in float
+        causal_mask = torch.triu(torch.full((seqlen, seqlen), -10000.0, device=scores.device), 1)
+        # TD [2022-09-30]: Adding is faster than masked_fill_ (idk why, just better kernel I guess)
+        scores = scores + causal_mask.to(dtype=scores.dtype)
+    attention = torch.softmax(scores, dim=-1)
+    attention_drop = F.dropout(attention, dropout_p)
+    output = torch.einsum('bhts,bshd->bthd', attention_drop , v)
+    return output.to(dtype=qkv.dtype)
+
+def flops(batch, seqlen, headdim, nheads, causal, mode="fwd"):
+    assert mode in ["fwd", "bwd", "fwd_bwd"]
+    f = 4 * batch * seqlen**2 * nheads * headdim // (2 if causal else 1)
+    return f if mode == "fwd" else (2.5 * f if mode == "bwd" else 3.5 * f)
+
+def efficiency(flop, time):
+    return (flop / time / 10**12) if not math.isnan(time) else 0.0
+
+def time_fwd(func, *args, **kwargs):
+    time.sleep(1) # Sleep to avoid residual power throttling from the previous benchmark
+    time_f = benchmark_forward(func, *args, **kwargs)
+    return time_f[1].mean
+
+
+torch.manual_seed(0)
+
+repeats = 30
+device = 'cuda'
+# dtype = torch.float16
+dtype = torch.float8_e4m3fn
+
+# bs_seqlen_vals = [(32, 512), (16, 1024), (8, 2048), (4, 4224), (2, 8448), (1, 8448 * 2)]
+bs_seqlen_vals = [(32, 512), (16, 1024), (8, 2048), (4, 4096), (2, 8192), (1, 8192 * 2)]
+# bs_seqlen_vals = [(4, 4096), (2, 8192), (1, 8192 * 2)]
+# bs_seqlen_vals = [(32, 512), (16, 1024), (8, 2048)]
+causal_vals = [False, True]
+headdim_vals = [64, 128, 256]
+dim = 2048
+# dim = 256
+dropout_p = 0.0
+
+methods = (["Pytorch", "Flash3"]
+        + (["cuDNN"] if cudnn is not None else [])
+        # + (["Triton"] if attention_triton is not None else [])
+        #    + (["xformers.c"] if xops is not None else [])
+        #    + (["xformers.f"] if xops is not None else [])
+           )
+
+time_f = {}
+time_b = {}
+time_f_b = {}
+speed_f = {}
+speed_b = {}
+speed_f_b = {}
+for causal in causal_vals:
+    for headdim in headdim_vals:
+        for batch_size, seqlen in bs_seqlen_vals:
+            torch.cuda.empty_cache()
+            config = (causal, headdim, batch_size, seqlen)
+            nheads = dim // headdim
+            q, k, v = [torch.randn(batch_size, seqlen, nheads, headdim, device=device, dtype=torch.bfloat16, requires_grad=False) for _ in range(3)]
+            
+            qkv = torch.stack([q, k, v], dim=2)
+            qkv = qkv.to(torch.bfloat16)
+            f = time_fwd(attention_pytorch, qkv, dropout_p, causal=causal, repeats=repeats, verbose=False)
+            time_f[config, "Pytorch"] = f
+            res_baseline = attention_pytorch(qkv, dropout_p, causal=causal)
+
+            if attention_triton is not None:
+                q_transposed = q.transpose(1, 2).contiguous().to(torch.float8_e4m3fn)
+                k_transposed = k.transpose(1, 2).contiguous().to(torch.float8_e4m3fn)
+                v_transposed = v.transpose(1, 2).contiguous().permute(0, 1, 3, 2).to(torch.float8_e4m3fn)
+                scale = 1 / math.sqrt(headdim)
+                f = time_fwd(
+                    attention_triton, q_transposed, k_transposed, v_transposed,
+                    causal, scale, repeats=5, verbose=False, desc='Triton'
+                )
+                f = time_fwd(
+                    attention_triton, q_transposed, k_transposed, v_transposed,
+                    causal, scale, repeats=repeats, verbose=False, desc='Triton'
+                )
+                time_f[config, "Triton"] = f
+                res = attention_triton(
+                    q_transposed, k_transposed, v_transposed.permute(0, 1, 3, 2),
+                    causal, scale
+                ).half().transpose(1, 2)
+                torch.testing.assert_close(res, res_baseline, atol=0.5, rtol=0.5)
+
+            # out = torch.empty_like(q)
+            q, k, v = q.to(dtype), k.to(dtype), v.to(dtype)
+            softmax_scale = q.shape[-1] ** (-0.5)
+            descale_q = torch.tensor([1.0], dtype=torch.float32, device='cuda')
+            descale_k = torch.tensor([1.0], dtype=torch.float32, device='cuda')
+            descale_v = torch.tensor([1.0], dtype=torch.float32, device='cuda')
+
+            # f = time_fwd(flash_attn_func, q, k, v, causal=causal, repeats=repeats, verbose=False)
+            f = time_fwd(
+                _flash_attn_forward,
+                q, 
+                k, 
+                v, 
+                softmax_scale, 
+                causal=causal,
+                window_size=(-1,-1),
+                descale_q=descale_q, 
+                descale_k=descale_k, 
+                descale_v=descale_v, 
+                repeats=repeats, 
+                verbose=False
+            )
+
+            # res = flash_attn_func(q, k, v, causal=causal)
+            # torch.testing.assert_close(res.half(), res_baseline, atol=0.05, rtol=0.05)
+
+            time_f[config, "Flash3"] = f
+
+            if cudnn is not None:
+                qkv_fp8 = qkv.to(dtype)
+                time.sleep(1) # Sleep to avoid residual power throttling from the previous benchmark
+                f = time_fwd(
+                    cudnn_spda_setup(
+                        qkv_fp8, seqlen, seqlen,
+                        causal=causal
+                    ),
+                    repeats=repeats, verbose=False
+                )
+                time_f[config, "cuDNN"] = f
+                # res, amax_o = cudnn_spda_setup(
+                #     qkv_fp8, seqlen, seqlen,
+                #     causal=causal
+                # )()
+                # res = res.half()
+                # TODO: CUDNN has numerics issues when
+                # num_heads=16, dim=128, seq_len=1024, batch_size=2
+                # or larger sizes.
+                # res_cpu = res.cpu().reshape(-1)
+                # res_baseline_cpu = res_baseline.cpu().reshape(-1)
+                # print(amax_o)
+                # print(res)
+                # print(res_baseline)
+                # for i in range(len(res_cpu)):
+                #     item = res_cpu[i]
+                #     item_baseline = res_baseline_cpu[i]
+                #     if abs(item - item_baseline) > 0.5:
+                #         print(i)
+                #         print(item)
+                #         print(item_baseline)
+                # torch.testing.assert_close(res, res_baseline, atol=0.05, rtol=0.05)
+
+            print(f"### causal={causal}, headdim={headdim}, batch_size={batch_size}, seqlen={seqlen} ###")
+            for method in methods:
+                speed_f[config, method] = efficiency(
+                    flops(batch_size, seqlen, headdim, nheads, causal, mode="fwd"),
+                    time_f[config, method]
+                )
+                #print (time_f[config,method])
+                print(
+                    f"{method} fwd: {speed_f[config, method]:.2f} TFLOPs/s, {time_f[config, method] * 1e3} ms, "
+                )
+
+
+# with open('flash3_attn_time.plk', 'wb') as fp:
+#     pickle.dump((time_f, time_b, time_f_b), fp, protocol=pickle.HIGHEST_PROTOCOL)

Code/Baselines/flash-attention/hopper/epilogue_fwd.hpp

@@ -0,0 +1,485 @@
+/******************************************************************************
+ * Copyright (c) 2024, Jay Shah, Ganesh Bikshandi, Ying Zhang, Vijay Thakkar, Pradeep Ramani, Tri Dao.
+ ******************************************************************************/
+
+#pragma once
+
+#include <cutlass/cutlass.h>
+#include <cutlass/fast_math.h>  // For FastDivMod
+#include "cute/tensor.hpp"
+
+#include "cutlass/gemm/collective/builders/sm90_common.inl"
+#include "cutlass/epilogue/collective/builders/sm90_common.inl"
+
+#include "seqlen.h"
+#include "named_barrier.hpp"
+#include "pack_gqa.h"
+#include "utils.h"
+
+namespace flash {
+
+using namespace cute;
+
+template <class TileShape_MNK_PV_, class ClusterShape_, class Element_, class ArchTag_,
+          int NumEpilogueThreads_, bool Varlen_, bool PackGQA_, bool Split_, bool FP8PermuteCol=false>
+struct CollectiveEpilogueFwd {
+
+    using TileShape_MNK_PV = TileShape_MNK_PV_;
+    using ClusterShape = ClusterShape_;
+    using Element = Element_;
+    using ElementPartial = float;
+    using ArchTag = ArchTag_;
+    static constexpr int NumEpilogueThreads = NumEpilogueThreads_;
+    static constexpr bool Varlen = Varlen_;
+    static constexpr bool PackGQA = PackGQA_;
+    static constexpr bool Split = Split_;
+    static constexpr bool Use_smem = !(Split && !Varlen);
+    static constexpr bool Use_TMA_O = ArchTag::kMinComputeCapability >= 90 && !Varlen && !Split && !PackGQA;
+
+    static_assert(ArchTag::kMinComputeCapability >= 80);
+    static_assert(ArchTag::kMinComputeCapability >= 90 || CUTE_STATIC_V(size(ClusterShape{})) == 1);
+    static_assert(sizeof(Element) <= 2);
+
+    static constexpr int kBlockM = get<0>(TileShape_MNK_PV{});
+    static constexpr int kHeadDimV = get<1>(TileShape_MNK_PV{});
+
+    static constexpr bool LargeHeadDimV = kHeadDimV > 256;
+
+    using GmemTiledCopyOTMA = cute::SM90_TMA_STORE;
+
+    // These are for storing the output tensor without TMA (e.g., for setting output to zero)
+    static constexpr int kGmemElemsPerStore = sizeof(cute::uint128_t) / sizeof(Element);
+    static_assert(kHeadDimV % kGmemElemsPerStore == 0, "Headdim must be a multiple of kGmemElemsPerStore");
+    // We want each "row" to have 64 elements (128 bytes, i.e. 1 cache line). We want each thread to have 4 elements
+    // in the M direction and 2 elements in the K direction. In the case of PackGQA, this reduces the number of times
+    // we need to call divmod.
+    static constexpr int kBytePerRow = kHeadDimV * sizeof(Element);
+    static constexpr int kBlockKGmem = (kBytePerRow % 128 == 0 ? 128 : (kBytePerRow % 64 == 0 ? 64 : 32)) / sizeof(Element);
+    static constexpr int kGmemThreadsPerRow = kBlockKGmem / kGmemElemsPerStore;
+    // If PackGQA, we split the work of compute O_ptr among threads in the same row, so we need this to within a warp
+    static_assert(cutlass::NumThreadsPerWarp % kGmemThreadsPerRow == 0);
+    static_assert(NumEpilogueThreads % kGmemThreadsPerRow == 0, "NumEpilogueThreads must be a multiple of kGmemThreadsPerRow");
+    using GmemLayoutAtom = Layout<Shape <Int<NumEpilogueThreads / kGmemThreadsPerRow>, Int<kGmemThreadsPerRow>>,
+                                  Stride<Int<kGmemThreadsPerRow>, _1>>;
+    static_assert(kBlockM % CUTE_STATIC_V(shape<0>(GmemLayoutAtom{})) == 0, "kBlockM must be a multiple of NumEpilogueThreads / kGmemThreadsPerRow");
+    using GmemTiledCopyO = decltype(
+        make_tiled_copy(Copy_Atom<AutoVectorizingCopyWithAssumedAlignment<128>, Element>{},
+                        GmemLayoutAtom{},
+                        Layout<Shape<_1, Int<kGmemElemsPerStore>>>{}));  // Val layout, 8 or 16 vals per store
+
+    using SmemLayoutAtomOTMA = decltype(cutlass::gemm::collective::detail::ss_smem_selector<GMMA::Major::K, Element,
+        decltype(cute::get<0>(TileShape_MNK_PV{})), decltype(cute::get<1>(TileShape_MNK_PV{}))>());
+    using SmemLayoutOTMA = decltype(tile_to_shape(SmemLayoutAtomOTMA{}, select<0, 1>(TileShape_MNK_PV{})));
+    static constexpr int kSwizzle = kBlockKGmem == 128 ? 4 : (kBlockKGmem == 64 ? 3 : (kBlockKGmem == 32 ? 2 : 1));
+    static constexpr int kSwizzleBase = sizeof(Element) == 4 ? 2 : (sizeof(Element) == 2 ? 3 : 4);
+    using SmemLayoutAtomO = decltype(
+        composition(Swizzle<kSwizzle, kSwizzleBase, kSwizzleBase>{},
+                    Layout<Shape<_8, Int<kBlockKGmem>>,
+                           Stride<Int<kBlockKGmem>, _1>>{}));
+    using SmemLayoutOSTS = decltype(tile_to_shape(SmemLayoutAtomO{}, select<0, 1>(TileShape_MNK_PV{})));
+    using SmemLayoutO = std::conditional_t<ArchTag::kMinComputeCapability >= 90, SmemLayoutOTMA, SmemLayoutOSTS>;
+
+    using ShapeO = cute::Shape<int32_t, int32_t, int32_t, int32_t, int32_t>;  // (seqlen_q, d, head, batch, num_splits)
+    using StrideO = cute::Stride<int64_t, _1, int64_t, int64_t, int64_t>;
+    using StrideLSE = cute::Stride<_1, int64_t, int64_t, int64_t>;            // (seqlen_q, head, batch, num_splits)
+    // ((qhead_per_khead, seqlen_q), d, nheads_kv, batch, num_splits)
+    using ShapeOPacked = std::conditional_t<!PackGQA, ShapeO, cute::Shape<cute::Shape<int32_t, int32_t>, int32_t, int32_t, int32_t, int32_t>>;
+    using StrideOPacked = std::conditional_t<!PackGQA, StrideO, cute::Stride<cute::Stride<int64_t, int64_t>, _1, int64_t, int64_t, int64_t>>;
+    // ((qhead_per_khead, seqlen_q), nheads_kv, batch, num_splits)
+    using ShapeLSEPacked = std::conditional_t<!PackGQA, cute::Shape<int32_t, int32_t, int32_t, int32_t>, cute::Shape<cute::Shape<int32_t, int32_t>, int32_t, int32_t, int32_t>>;
+    using StrideLSEPacked = std::conditional_t<!PackGQA, StrideLSE, cute::Stride<cute::Stride<int64_t, _1>, int64_t, int64_t, int64_t>>;
+
+    using EpilogueTile_MN = decltype(select<0, 1>(TileShape_MNK_PV{}));
+    using CopyOpR2S = std::conditional_t<
+        ArchTag::kMinComputeCapability >= 90,
+        // cute::SM90_U32x4_STSM_N if Element size is 2 bytes (fp16, bf16)
+        decltype(cutlass::epilogue::collective::detail::sm90_get_smem_store_op_for_accumulator<StrideO, Element, EpilogueTile_MN>()),
+        AutoVectorizingCopyWithAssumedAlignment<128>
+    >;
+    using SmemCopyAtomO = Copy_Atom<CopyOpR2S, Element>;
+
+    // static constexpr size_t SmemAlignmentO = cutlass::detail::alignment_for_swizzle(SmemLayoutO{});
+    // static_assert(SmemAlignmentO >= 128, "Require at least 128B alignment");
+    // struct TensorStorage : cute::aligned_struct<SmemAlignmentO> {
+    //     cute::array_aligned<Element, Use_smem ? cute::cosize_v<SmemLayoutO> : 0, SmemAlignmentO> smem_o;
+    // };
+    struct TensorStorage : cute::aligned_struct<128> {
+        cute::array_aligned<Element, Use_smem ? cute::cosize_v<SmemLayoutO> : 0> smem_o;
+    };
+
+    using TMA_O = std::conditional_t<
+        Use_TMA_O,
+        decltype(make_tma_copy(
+            GmemTiledCopyOTMA{},
+            make_tensor(make_gmem_ptr(static_cast<Element*>(nullptr)), ShapeO{}, StrideO{}),
+            SmemLayoutOTMA{},
+            select<0, 1>(TileShape_MNK_PV{}),
+            _1{})),  // no mcast for O
+        std::nullptr_t
+    >;
+
+    // Host side kernel arguments
+    struct Arguments {
+        Element* ptr_O;
+        ShapeO const shape_O;
+        StrideO const stride_O;
+        ElementPartial* ptr_O_partial;
+        StrideO const stride_O_partial;
+        float* ptr_LSE;
+        StrideLSE const stride_LSE;
+        float* ptr_LSE_partial;
+        StrideLSE const stride_LSE_partial;
+        int32_t const nheads_kv;
+        int const* cu_seqlens = nullptr;
+        int const* seqused = nullptr;
+    };
+
+    // Device side kernel params
+    struct Params {
+        Element* ptr_O;
+        ShapeO const shape_O;
+        StrideO const stride_O;
+        ShapeOPacked const shape_O_packed;
+        StrideOPacked const stride_O_packed;
+        ElementPartial* ptr_O_partial;
+        StrideO const stride_O_partial;
+        StrideOPacked const stride_O_partial_packed;
+        float* ptr_LSE;
+        StrideLSE const stride_LSE;
+        ShapeLSEPacked const shape_LSE_packed;
+        StrideLSEPacked const stride_LSE_packed;
+        float* ptr_LSE_partial;
+        StrideLSE const stride_LSE_partial;
+        StrideLSEPacked const stride_LSE_partial_packed;
+        cutlass::FastDivmod qhead_per_khead_divmod;
+        TMA_O tma_store_O;
+        int const* cu_seqlens = nullptr;
+        int const* seqused = nullptr;
+    };
+
+    static Params
+    to_underlying_arguments(Arguments const& args) {
+        Tensor mO = make_tensor(make_gmem_ptr(args.ptr_O), args.shape_O, args.stride_O);
+        TMA_O tma_store_O = [&]{
+            if constexpr (Use_TMA_O) {
+                return make_tma_copy(GmemTiledCopyOTMA{}, mO, SmemLayoutO{}, select<0, 1>(TileShape_MNK_PV{}), _1{}); // no mcast
+            } else {
+                return nullptr;
+            }
+        }();
+        // If PackGQA, reshape O to be ((qhead_per_khead, seqlen_q), head_size, nhead_k, batch_size, num_splits)
+        int const qhead_per_khead = !PackGQA ? 1 : cute::ceil_div(get<2>(args.shape_O), args.nheads_kv);
+        auto const shape_O_packed = cute::conditional_return<!PackGQA>(
+            args.shape_O,
+            make_shape(make_shape(qhead_per_khead, get<0>(args.shape_O)), get<1>(args.shape_O), args.nheads_kv, get<3>(args.shape_O), get<4>(args.shape_O))
+        );
+        auto const stride_O_packed = cute::conditional_return<!PackGQA>(
+            args.stride_O,
+            make_stride(make_stride(get<2>(args.stride_O), get<0>(args.stride_O)), get<1>(args.stride_O), get<2>(args.stride_O) * qhead_per_khead, get<3>(args.stride_O), get<4>(args.stride_O))
+        );
+        auto const stride_O_partial_packed = cute::conditional_return<!PackGQA>(
+            args.stride_O_partial,
+            make_stride(make_stride(get<2>(args.stride_O_partial), get<0>(args.stride_O_partial)), get<1>(args.stride_O_partial), get<2>(args.stride_O_partial) * qhead_per_khead, get<3>(args.stride_O_partial), get<4>(args.stride_O_partial))
+        );
+        // If PackGQA, Reshape LSE to be ((qhead_per_khead, seqlen_q), nhead_k, batch_size, num_splits)
+        auto const shape_LSE_packed = cute::conditional_return<!PackGQA>(
+            select<0, 2, 3, 4>(args.shape_O),
+            make_shape(make_shape(qhead_per_khead, get<0>(args.shape_O)), args.nheads_kv, get<3>(args.shape_O), get<4>(args.shape_O))
+        );
+        auto const stride_LSE_packed = cute::conditional_return<!PackGQA>(
+            args.stride_LSE,
+            make_stride(make_stride(get<1>(args.stride_LSE), get<0>(args.stride_LSE)), get<1>(args.stride_LSE) * qhead_per_khead, get<2>(args.stride_LSE), get<3>(args.stride_LSE))
+        );
+        auto const stride_LSE_partial_packed = cute::conditional_return<!PackGQA>(
+            args.stride_LSE_partial,
+            make_stride(make_stride(get<1>(args.stride_LSE_partial), get<0>(args.stride_LSE_partial)), get<1>(args.stride_LSE_partial) * qhead_per_khead, get<2>(args.stride_LSE_partial), get<3>(args.stride_LSE_partial))
+        );
+        return {args.ptr_O, args.shape_O, args.stride_O, shape_O_packed, stride_O_packed,
+                args.ptr_O_partial, args.stride_O_partial, stride_O_partial_packed,
+                args.ptr_LSE, args.stride_LSE, shape_LSE_packed, stride_LSE_packed,
+                args.ptr_LSE_partial, args.stride_LSE_partial, stride_LSE_partial_packed,
+                cutlass::FastDivmod(qhead_per_khead),
+                tma_store_O, args.cu_seqlens, args.seqused};
+    }
+
+    /// Issue Tma Descriptor Prefetch -- ideally from a single thread for best performance
+    CUTLASS_DEVICE
+    static void prefetch_tma_descriptors(Params const& params) {
+        if constexpr (Use_TMA_O) {
+            cute::prefetch_tma_descriptor(params.tma_store_O.get_tma_descriptor());
+        }
+    }
+
+    template <typename SharedStorage, typename FrgTensorO, typename FrgTensorLSE, typename TiledMma>
+    CUTLASS_DEVICE void
+    store(Params const& params,
+          FrgTensorO& tOrO,
+          FrgTensorLSE const& lse,
+          SharedStorage& shared_storage,
+          TiledMma tiled_mma,
+          int thread_idx,
+          cute::tuple<int32_t, int32_t, int32_t, int32_t> const& block_coord
+          ) {
+
+        auto [m_block, bidh, bidb, split_idx] = block_coord;
+        int num_splits = get<4>(params.shape_O_packed);
+        if constexpr (Split && Varlen) {
+            uint32_t num_splits_dynamic_u = reinterpret_cast<uint32_t const&>(split_idx) >> 16; // first 16 bits are for num_splits
+            int num_splits_dynamic = reinterpret_cast<int&>(num_splits_dynamic_u);
+            num_splits = num_splits_dynamic > 0 ? num_splits_dynamic : num_splits;
+            split_idx &= 0x0000FFFF;  // Only use the lower 16 bits of split_idx
+        }
+        bool const is_split = !Split ? false : (!Varlen ? true : num_splits > 1);
+
+        Tensor sO = make_tensor(make_smem_ptr(shared_storage.tensors.epilogue.smem_o.data()), SmemLayoutO{});
+        // Tensor sO_pi = cute::as_position_independent_swizzle_tensor(sO);
+
+        static constexpr bool NeedFP8Permute = FP8PermuteCol && (sizeof(Element) == 2 || sizeof(Element) == 4);
+        // If we will possibly need tOrO in FP32, we'd want to permute tOrO before type conversion.
+        // Otherwise we can permute after conversion.
+        if constexpr (NeedFP8Permute && Split) { flash::permute_output_fp8_Vcolmajor(tOrO); }
+        Tensor tOrO_out = make_tensor_like<Element>(tOrO);
+        flash::convert_type_out(tOrO, tOrO_out);
+        if constexpr (NeedFP8Permute && !Split) { flash::permute_output_fp8_Vcolmajor(tOrO_out); }
+
+        // Make sure all WGs have finished reading V
+        // Technically we don't need this if we're not using smem, but the mainloop makes the assumption that
+        // all epilogue threads sync at least once during the epilogue (so that we can start loading Q with
+        // cp.async if we need).
+        flash::named_barrier_sync(NumEpilogueThreads, cutlass::arch::ReservedNamedBarriers::EpilogueBarrier);
+
+        // Step 1: Write O from rmem -> smem
+        if constexpr (Use_smem) {
+            auto smem_tiled_copy_O = make_tiled_copy_C(SmemCopyAtomO{}, tiled_mma);
+            auto smem_thr_copy_O = smem_tiled_copy_O.get_thread_slice(thread_idx);
+            Tensor taccOrO = smem_thr_copy_O.retile_S(tOrO_out);        // ((Atom,AtomNum), MMA_M, MMA_N)
+            Tensor taccOsO = smem_thr_copy_O.partition_D(sO);     // ((Atom,AtomNum),PIPE_M,PIPE_N)
+            // Tensor taccOsO = smem_thr_copy_O.partition_D(sO_pi);     // ((Atom,AtomNum),PIPE_M,PIPE_N)
+            cute::copy(smem_tiled_copy_O, taccOrO, taccOsO);
+            if constexpr (Use_TMA_O) {
+                cutlass::arch::fence_view_async_shared(); // ensure smem writes are visible to TMA
+                cutlass::arch::NamedBarrier::arrive(NumEpilogueThreads + cutlass::NumThreadsPerWarp,
+                                                    cutlass::arch::ReservedNamedBarriers::EpilogueBarrier);
+            } else {
+                flash::named_barrier_sync(NumEpilogueThreads, cutlass::arch::ReservedNamedBarriers::EpilogueBarrier);
+            }
+        } else {
+            if constexpr (ArchTag::kMinComputeCapability >= 90) {
+                #pragma unroll
+                for (uint32_t cta_id = 0; cta_id < size(ClusterShape{}); ++cta_id) {
+                    shared_storage.pipelines.barrier_O.arrive(cta_id);
+                }
+            }
+        }
+
+        flash::SeqlenInfo<Varlen, kBlockM> seqlen_info{bidb, size<0>(params.shape_O), params.cu_seqlens, params.seqused};
+        bool is_varlen = Varlen && params.cu_seqlens;
+        int offset_o = seqlen_info.offset;
+        int seqlen_o = seqlen_info.seqlen;
+        int warp_group_idx = __shfl_sync(0xFFFFFFFF, thread_idx / cutlass::NumThreadsPerWarpGroup, 0);
+
+        // Step 2: Write LSE from rmem -> gmem
+        auto thread_mma = tiled_mma.get_thread_slice(thread_idx);
+        // (MMA,MMA_M,MMA_K)
+        Tensor taccOcO = thread_mma.partition_C(cute::make_identity_tensor(select<0, 1>(TileShape_MNK_PV{})));
+        static_assert(decltype(size<0, 0>(taccOcO))::value == 2);
+        static_assert(decltype(size<0, 1>(taccOcO))::value == 2);
+        Tensor taccOcO_rowcol = make_tensor(taccOcO.data(), flash::convert_layout_acc_rowcol(taccOcO.layout()));
+        Tensor taccOcO_row = taccOcO_rowcol(_, _0{});
+        CUTE_STATIC_ASSERT_V(size(lse) == size(taccOcO_row));                     // MMA_M
+
+        using PackGQA_t = flash::PackGQAManager<get<0>(TileShape_MNK_PV{}), get<1>(TileShape_MNK_PV{}), NumEpilogueThreads, Element>;
+        using PackGQApartial_t = flash::PackGQAManager<get<0>(TileShape_MNK_PV{}), get<1>(TileShape_MNK_PV{}), NumEpilogueThreads, ElementPartial>;
+
+        Tensor mLSE = make_tensor(make_gmem_ptr((!is_split ? params.ptr_LSE : params.ptr_LSE_partial) + offset_o * get<0>(!is_split ? params.stride_LSE : params.stride_LSE_partial)),
+                                  params.shape_LSE_packed,
+                                  !is_split ? params.stride_LSE_packed : params.stride_LSE_partial_packed)(_, bidh, !is_varlen ? bidb : 0, !is_split ? 0 : split_idx);
+        // if (thread_idx == 0) { printf("Before LSE write, m_block: %d, bidh: %d, bidb: %d, split_idx: %d, offset_o: %d, seqlen_o: %d\n", m_block, bidh, bidb, split_idx, offset_o, seqlen_o); print(mLSE); printf("\n"); }
+        if (!LargeHeadDimV || warp_group_idx == 0) {
+            if constexpr (!PackGQA) {
+                #pragma unroll
+                for (int mi = 0; mi < size(lse); ++mi) {
+                    int const row = m_block * kBlockM + get<0>(taccOcO_row(mi));
+                    if (get<1>(taccOcO_row(_0{})) == 0 && row < seqlen_o) { mLSE(row) = lse(mi); }
+                }
+            } else {
+                PackGQA_t::store_LSE(mLSE, lse, tiled_mma, params.qhead_per_khead_divmod, thread_idx, seqlen_o, m_block);
+            }
+        }
+
+        // Step 3: Write O from smem -> gmem
+        if constexpr (Use_TMA_O) {
+            Tensor mO = params.tma_store_O.get_tma_tensor(params.shape_O)(_, _, bidh, bidb, split_idx);
+            Tensor gO = local_tile(mO, select<0, 1>(TileShape_MNK_PV{}), make_coord(m_block, _0{}));  // (M, K)
+            auto block_tma_O = params.tma_store_O.get_slice(_0{});
+            Tensor tOgO = block_tma_O.partition_D(gO);  // (TMA, TMA_M, TMA_K)
+            Tensor tOsO = block_tma_O.partition_S(sO); // (TMA, TMA_M, TMA_K)
+            int warp_idx_sync = __shfl_sync(0xffffffff, thread_idx / cutlass::NumThreadsPerWarp, 0);
+            if (warp_idx_sync == NumEpilogueThreads / cutlass::NumThreadsPerWarp - 1) {
+                cutlass::arch::NamedBarrier::sync(NumEpilogueThreads + cutlass::NumThreadsPerWarp,
+                                                  cutlass::arch::ReservedNamedBarriers::EpilogueBarrier);
+                if (cute::elect_one_sync()) {
+                    cute::copy(params.tma_store_O, tOsO, tOgO);
+                    tma_store_arrive();
+                    tma_store_wait<0>();
+                    #pragma unroll
+                    for (uint32_t cta_id = 0; cta_id < size(ClusterShape{}); ++cta_id) {
+                        shared_storage.pipelines.barrier_O.arrive(cta_id);
+                    }
+                }
+            }
+        } else {  // Don't use TMA in Varlen case since we don't want to overwrite the output of another sequence
+            if (!is_split) {
+                Tensor mO = make_tensor(make_gmem_ptr(params.ptr_O + offset_o * get<0>(params.stride_O)), params.shape_O_packed, params.stride_O_packed)(_, _, bidh, !is_varlen ? bidb : 0, _0{});
+                Tensor gO = local_tile(mO, select<0, 1>(TileShape_MNK_PV{}), make_coord(m_block, _0{}));  // (M, K)
+                // if (thread_idx == 0) { printf("Before O write, m_block: %d, bidh: %d, bidb: %d, split_idx: %d, offset_o: %d, seqlen_o: %d, mO_addr = %p, addr diff = %d\n", m_block, bidh, bidb, split_idx, offset_o, seqlen_o, mO.data(), reinterpret_cast<int>(&mO(0)) - reinterpret_cast<int>(params.ptr_O)); }
+                GmemTiledCopyO gmem_tiled_copy_O;
+                auto gmem_thr_copy_O = gmem_tiled_copy_O.get_thread_slice(thread_idx);
+                Tensor tOsO = gmem_thr_copy_O.partition_S(sO);        // ((Atom,AtomNum),ATOM_M,ATOM_N)
+                // Tensor tOsO = gmem_thr_copy_O.partition_S(sO_pi);        // ((Atom,AtomNum),ATOM_M,ATOM_N)
+                Tensor tOrO = make_fragment_like(tOsO);
+                cute::copy(gmem_tiled_copy_O, tOsO, tOrO);
+                if constexpr (ArchTag::kMinComputeCapability >= 90) {
+                    cutlass::arch::fence_view_async_shared(); // ensure smem reads are done before next TMA to smem_v
+                    #pragma unroll
+                    for (uint32_t cta_id = 0; cta_id < size(ClusterShape{}); ++cta_id) {
+                        shared_storage.pipelines.barrier_O.arrive(cta_id);
+                    }
+                }
+                if constexpr (!PackGQA) {
+                    // (BLK_M,BLK_K) -> (blk_m,blk_k)
+                    Tensor tOcO = gmem_thr_copy_O.partition_D(cute::make_identity_tensor(select<0, 1>(TileShape_MNK_PV{})));
+                    Tensor tOpO = make_tensor<bool>(make_shape(size<2>(tOsO)));
+                    #pragma unroll
+                    for (int k = 0; k < size(tOpO); ++k) { tOpO(k) = get<1>(tOcO(_0{}, _0{}, k)) < get<1>(params.shape_O); }
+                    Tensor tOgO = gmem_thr_copy_O.partition_D(gO);
+                    // Clear_OOB_K must be false since we don't want to write zeros to gmem
+                    flash::copy</*Is_even_MN=*/false, /*Is_even_K=*/false, /*Clear_OOB_MN=*/false, /*Clear_OOB_K=*/false>(
+                        gmem_tiled_copy_O, tOrO, tOgO, tOcO, tOpO, seqlen_o - m_block * kBlockM
+                    );
+                } else {
+                    // If PackGQA, we split the work of compute O_ptr among threads in the same row
+                    PackGQA_t::store_O(mO, tOrO, params.qhead_per_khead_divmod, thread_idx, seqlen_o, m_block);
+                }
+            } else {
+                Tensor mOpartial = make_tensor(make_gmem_ptr(params.ptr_O_partial + offset_o * get<0>(params.stride_O_partial)), params.shape_O_packed, params.stride_O_partial_packed)(_, _, bidh, !is_varlen ? bidb : 0, split_idx);
+                Tensor gOpartial = local_tile(mOpartial, select<0, 1>(TileShape_MNK_PV{}), make_coord(m_block, _0{}));  // (M, K)
+                // We already arrived on barrier_O earlier if !Use_smem
+                if constexpr (Use_smem) {
+                    if constexpr (ArchTag::kMinComputeCapability >= 90) {
+                        #pragma unroll
+                        for (uint32_t cta_id = 0; cta_id < size(ClusterShape{}); ++cta_id) {
+                            shared_storage.pipelines.barrier_O.arrive(cta_id);
+                        }
+                    }
+                }
+                if constexpr (!PackGQA) {
+                    static constexpr int kGmemElemsPerStoreDirect = 2;
+                    cute::Copy_Atom<AutoVectorizingCopyWithAssumedAlignment<128>, ElementPartial> gmem_copy_direct;
+                    // Reshape acc from ((2, 2, V), MMA_M, MMA_N) to (nrow=(2, MMA_M), ncol=(2, V, MMA_N))
+                    Tensor tOrO_rowcol = make_tensor(tOrO.data(), flash::convert_layout_acc_rowcol(tOrO.layout()));
+                    Tensor tOrO_copy = cute::tiled_divide(tOrO_rowcol, Shape<_1, Int<kGmemElemsPerStoreDirect>>{});
+                    Tensor tOgO = thread_mma.partition_C(gOpartial);
+                    Tensor tOgO_rowcol = make_tensor(tOgO.data(), flash::convert_layout_acc_rowcol(tOgO.layout()));
+                    Tensor tOgO_copy = cute::tiled_divide(tOgO_rowcol, Shape<_1, Int<kGmemElemsPerStoreDirect>>{});
+                    Tensor taccOcO_col = taccOcO_rowcol(_0{}, _);
+                    #pragma unroll
+                    for (int m = 0; m < size(taccOcO_row); ++m) {
+                        if (get<0>(taccOcO_row(m)) < seqlen_o - m_block * kBlockM) {
+                            #pragma unroll
+                            for (int k = 0; k < size(taccOcO_col) / kGmemElemsPerStoreDirect; ++k) {
+                                if (get<1>(taccOcO_col(k * kGmemElemsPerStoreDirect)) < get<1>(params.shape_O)) {
+                                    cute::copy(gmem_copy_direct, tOrO_copy(_, m, k), tOgO_copy(_, m, k));
+                                }
+                            }
+                        }
+                    }
+                } else {
+                    PackGQApartial_t::store_O_direct(mOpartial, tOrO, tiled_mma, params.qhead_per_khead_divmod, thread_idx, seqlen_o, m_block);
+                }
+            }
+        }
+    }
+
+    CUTLASS_DEVICE void
+    store_tail() {
+        // Don't need to do tma_store_wait<0>() here since we already did in @store
+    }
+
+    // Write 0 to output and -inf to LSE
+    CUTLASS_DEVICE void
+    store_zero(
+         Params const& params,
+         int thread_idx,
+         cute::tuple<int32_t, int32_t, int32_t, int32_t> const& block_coord
+         ) {
+        static constexpr int kBlockM = get<0>(TileShape_MNK_PV{});
+        auto [m_block, bidh, bidb, split_idx] = block_coord;
+        int num_splits = get<4>(params.shape_O_packed);
+        if constexpr (Split && Varlen) {
+            uint32_t num_splits_dynamic_u = reinterpret_cast<uint32_t const&>(split_idx) >> 16; // first 16 bits are for num_splits
+            int num_splits_dynamic = reinterpret_cast<int&>(num_splits_dynamic_u);
+            num_splits = num_splits_dynamic > 0 ? num_splits_dynamic : num_splits;
+            split_idx &= 0x0000FFFF;  // Only use the lower 16 bits of split_idx
+        }
+        bool const is_split = !Split ? false : (!Varlen ? true : num_splits > 1);
+
+        flash::SeqlenInfo<Varlen, kBlockM> seqlen_info{bidb, size<0>(params.shape_O), params.cu_seqlens, params.seqused};
+        bool const is_varlen = Varlen && params.cu_seqlens;
+        int offset_o = seqlen_info.offset;
+        int seqlen_o = seqlen_info.seqlen;
+        int qhead_per_khead = !PackGQA ? 1 : params.qhead_per_khead_divmod.divisor;
+        Tensor mLSE = make_tensor(make_gmem_ptr((!is_split ? params.ptr_LSE : params.ptr_LSE_partial) + offset_o * get<0>(!is_split ? params.stride_LSE : params.stride_LSE_partial)),
+                                  params.shape_LSE_packed,
+                                  !is_split ? params.stride_LSE_packed : params.stride_LSE_partial_packed)(_, bidh, !is_varlen ? bidb : 0, !is_split ? 0 : split_idx);
+        Tensor gLSE = local_tile(mLSE, Shape<Int<kBlockM>>{}, make_coord(m_block));
+
+        static_assert(kBlockM <= NumEpilogueThreads);
+        if (thread_idx < kBlockM) {
+            const int row = m_block * kBlockM + thread_idx;
+            if constexpr (!PackGQA) {
+                if (row < seqlen_o) { mLSE(row) = -INFINITY; }
+            } else {
+                if (row < seqlen_o * qhead_per_khead) {
+                    int m_idx, h_idx;
+                    m_idx = params.qhead_per_khead_divmod.divmod(h_idx, row);
+                    // mLSE has shape ((qhead_per_khead, seqlen_q)) and it's unhappy with just 1 "make_coord"
+                    mLSE(make_coord(make_coord(h_idx, m_idx))) = -INFINITY;
+                }
+            }
+        }
+
+        // If split, we don't have to write 0 to mOpartial if the mha_combine kernel is used,
+        // since it will not use the value of O if LSE is -inf.
+        if (!is_split) {
+            Tensor mO = make_tensor(make_gmem_ptr(params.ptr_O + offset_o * get<0>(params.stride_O)), params.shape_O_packed, params.stride_O_packed)(_, _, bidh, !is_varlen ? bidb : 0, _0{});
+
+            GmemTiledCopyO gmem_tiled_copy_O;
+            auto gmem_thr_copy_O = gmem_tiled_copy_O.get_thread_slice(thread_idx);
+            Tensor tOcO = gmem_thr_copy_O.partition_D(cute::make_identity_tensor(select<0, 1>(TileShape_MNK_PV{})));
+            if constexpr (!PackGQA) {
+                Tensor tOpO = make_tensor<bool>(make_shape(size<2>(tOcO)));
+                #pragma unroll
+                for (int k = 0; k < size(tOpO); ++k) { tOpO(k) = get<1>(tOcO(_0{}, _0{}, k)) < get<1>(params.shape_O); }
+                Tensor gO = local_tile(mO, select<0, 1>(TileShape_MNK_PV{}), make_coord(m_block, _0{}));  // (M, K)
+                Tensor tOgO = gmem_thr_copy_O.partition_D(gO);
+                Tensor tOrO = make_fragment_like(tOgO);
+                cute::clear(tOrO);
+                // Clear_OOB_K must be false since we don't want to write zeros to gmem
+                flash::copy</*Is_even_MN=*/false, /*Is_even_K=*/false, /*Clear_OOB_MN=*/false, /*Clear_OOB_K=*/false>(
+                    gmem_tiled_copy_O, tOrO, tOgO, tOcO, tOpO, seqlen_o - m_block * kBlockM
+                );
+            } else {
+                // If PackGQA, we split the work of compute O_ptr among threads in the same row
+                using PackGQA_t = flash::PackGQAManager<get<0>(TileShape_MNK_PV{}), get<1>(TileShape_MNK_PV{}), NumEpilogueThreads, Element>;
+                Tensor tOrO = make_tensor<Element>(make_shape(Shape<_1, Int<kGmemElemsPerStore>>{}, size<1>(tOcO), size<2>(tOcO)));
+                cute::clear(tOrO);
+                PackGQA_t::store_O(mO, tOrO, params.qhead_per_khead_divmod, thread_idx, seqlen_o, m_block);
+            }
+        }
+
+    }
+
+};
+
+} // namespace flash

Code/Baselines/flash-attention/hopper/flash_bwd_kernel_sm80.h

@@ -0,0 +1,173 @@
+/******************************************************************************
+ * Copyright (c) 2024, Tri Dao.
+ ******************************************************************************/
+
+#pragma once
+
+#include "cute/tensor.hpp"
+
+#include <cutlass/cutlass.h>
+#include <cutlass/array.h>
+#include <cutlass/numeric_types.h>
+#include <cutlass/kernel_hardware_info.h>
+
+#include "utils.h"
+
+namespace flash {
+
+using namespace cute;
+
+template <class CollectiveMainloop_, class CollectiveEpilogue_, class TileScheduler_>
+class FlashAttnBwdSm80 {
+
+public:
+
+    // Type Aliases
+    static constexpr bool Is_causal = CollectiveMainloop_::Is_causal;
+    static constexpr bool Is_local = CollectiveMainloop_::Is_local;
+    static_assert(CollectiveMainloop_::Varlen == CollectiveEpilogue_::Varlen);
+    static constexpr bool Varlen = CollectiveMainloop_::Varlen;
+
+    // Mainloop derived types
+    using CollectiveMainloop = CollectiveMainloop_;
+    using TileShape_MNK = typename CollectiveMainloop::TileShape_MNK;
+    using TiledMmaSdP = typename CollectiveMainloop::TiledMmaSdP;
+    using TiledMmadKV = typename CollectiveMainloop::TiledMmadKV;
+    using ArchTag = typename CollectiveMainloop::ArchTag;
+    using MainloopArguments = typename CollectiveMainloop::Arguments;
+    using MainloopParams = typename CollectiveMainloop::Params;
+    static constexpr bool dKV_swapAB = CollectiveMainloop::dKV_swapAB;
+
+    // Epilogue derived types
+    using CollectiveEpilogue = CollectiveEpilogue_;
+    using EpilogueArguments = typename CollectiveEpilogue::Arguments;
+    using EpilogueParams = typename CollectiveEpilogue::Params;
+
+    static_assert(ArchTag::kMinComputeCapability >= 80);
+
+    using TileScheduler = TileScheduler_;
+    using TileSchedulerArguments = typename flash::TileSchedulerArguments;
+    using TileSchedulerParams = typename TileScheduler::Params;
+
+    static constexpr uint32_t NumThreads = CUTE_STATIC_V(size(TiledMmaSdP{}));
+    static constexpr uint32_t MaxThreadsPerBlock = CUTE_STATIC_V(size(TiledMmaSdP{}));
+    static constexpr uint32_t MinBlocksPerMultiprocessor = 1;
+
+    // Kernel level shared memory storage
+    struct SharedStorage {
+        struct TensorStorage : cute::aligned_struct<128> {
+            union {
+                typename CollectiveMainloop::TensorStorage mainloop;
+                typename CollectiveEpilogue::TensorStorage epilogue;
+            };
+        } tensors;
+
+        alignas(16) typename TileScheduler::SharedStorage smem_scheduler;
+
+    };
+
+    static constexpr int SharedStorageSize = sizeof(SharedStorage);
+
+    // Device side arguments
+    struct Arguments {
+        MainloopArguments mainloop{};
+        EpilogueArguments epilogue{};
+        cutlass::KernelHardwareInfo hw_info{};
+        TileSchedulerArguments scheduler{};
+    };
+
+    // Kernel entry point API
+    struct Params {
+        MainloopParams mainloop{};
+        EpilogueParams epilogue{};
+        cutlass::KernelHardwareInfo hw_info{};
+        TileSchedulerParams scheduler{};
+    };
+
+    //
+    // Methods
+    //
+
+    // Convert to underlying arguments. In this case, a simple copy for the aliased type.
+    static
+    Params
+    to_underlying_arguments(Arguments const& args) {
+        CUTLASS_TRACE_HOST("to_underlying_arguments():");
+
+        // Get SM count if needed, otherwise use user supplied SM count
+        int sm_count = args.hw_info.sm_count;
+        if (sm_count <= 0) {
+            CUTLASS_TRACE_HOST("  WARNING: Arguments do not include a valid SM count.\n"
+                "  For optimal performance, populate the arguments KernelHardwareInfo struct with the SM count.");
+            sm_count = cutlass::KernelHardwareInfo::query_device_multiprocessor_count(args.hw_info.device_id);
+        }
+
+        CUTLASS_TRACE_HOST("to_underlying_arguments(): Setting persistent grid SM count to " << sm_count);
+
+        cutlass::KernelHardwareInfo hw_info{args.hw_info.device_id, sm_count};
+        return {
+            CollectiveMainloop::to_underlying_arguments(args.mainloop),
+            CollectiveEpilogue::to_underlying_arguments(args.epilogue),
+            hw_info,
+            TileScheduler::to_underlying_arguments(args.scheduler)
+        };
+    }
+
+    // Computes the kernel launch grid shape based on runtime parameters
+    static dim3
+    get_grid_shape(Params const& params) {
+        return TileScheduler::get_grid_shape(params.scheduler, params.hw_info.sm_count);
+    }
+
+    static dim3
+    get_block_shape() {
+        return dim3(MaxThreadsPerBlock, 1, 1);
+    }
+
+    CUTLASS_DEVICE
+    void
+    operator()(Params const& params, char* smem_buf) {
+
+        static constexpr int kBlockM = get<0>(TileShape_MNK{});
+        static constexpr int kBlockN = get<1>(TileShape_MNK{});
+
+        SharedStorage& shared_storage = *reinterpret_cast<SharedStorage*>(smem_buf);
+
+        CollectiveMainloop mainloop;
+        CollectiveEpilogue epilogue;
+
+        TileScheduler scheduler(reinterpret_cast<typename TileScheduler::SharedStorage*>(&shared_storage.smem_scheduler));
+        // Initialize matmul objects.
+        TiledMmadKV tiled_mma_dKV;
+
+        scheduler.init_consumer();
+
+        int warp_idx = cutlass::canonical_warp_idx_sync();
+        CUTLASS_PRAGMA_NO_UNROLL
+        for (auto work_tile_info = warp_idx == 0 ? scheduler.template get_initial_work</*IsProducerWarp=*/true>(params.scheduler) : scheduler.template get_initial_work</*IsProducerWarp=*/false>(params.scheduler);
+             work_tile_info.is_valid(params.scheduler);
+             work_tile_info = warp_idx == 0 ? scheduler.template get_next_work</*IsProducerWarp=*/true>(params.scheduler, work_tile_info) : scheduler.template get_next_work</*IsProducerWarp=*/false>(params.scheduler, work_tile_info)) {
+
+            auto block_coord_ = work_tile_info.get_block_coord(params.scheduler);
+            auto [n_block, bidh, bidb, _ /*split_idx*/] = block_coord_;
+            cute::tuple<int32_t, int32_t, int32_t> block_coord = {n_block, bidh, bidb};
+
+            // dK and dV output accumulator.
+            Tensor tdKrdK = partition_fragment_C(tiled_mma_dKV, select<!dKV_swapAB ? 1 : 2, !dKV_swapAB? 2 : 1>(TileShape_MNK{}));
+            Tensor tdVrdV = partition_fragment_C(tiled_mma_dKV, select<!dKV_swapAB ? 1 : 2, !dKV_swapAB? 2 : 1>(TileShape_MNK{}));
+            bool tile_valid = mainloop.mma(params.mainloop, tdKrdK, tdVrdV, threadIdx.x,
+                                           block_coord, shared_storage);
+            scheduler.prefetch_next_work(params.scheduler, work_tile_info);
+            if (tile_valid) {
+                epilogue.store(params.epilogue, tdKrdK, tdVrdV, shared_storage, tiled_mma_dKV,
+                               threadIdx.x, block_coord);
+            } else {
+                epilogue.store_zero(params.epilogue, threadIdx.x, block_coord);
+            }
+        }
+
+    }
+
+};
+
+} // namespace flash

Code/Baselines/flash-attention/hopper/flash_bwd_postprocess_kernel.h

@@ -0,0 +1,256 @@
+/******************************************************************************
+ * Copyright (c) 2024, Jay Shah, Ganesh Bikshandi, Ying Zhang, Vijay Thakkar, Pradeep Ramani, Tri Dao.
+ ******************************************************************************/
+
+#pragma once
+
+#include "cute/tensor.hpp"
+
+#include <cutlass/cutlass.h>
+#include <cutlass/array.h>
+#include <cutlass/numeric_types.h>
+#include <cutlass/numeric_conversion.h>
+#include "cutlass/arch/barrier.h"
+
+#include "seqlen.h"
+#include "utils.h"
+
+namespace flash {
+
+using namespace cute;
+
+template <class TileShape_MK_, class Element, class ElementAccum, class ArchTag_, int kNThreads, class TiledMma, bool dQ_swapAB>
+class FlashAttnBwdPostprocessConvertdQ {
+
+public:
+
+    // Type Aliases
+    using TileShape_MK = TileShape_MK_;
+    using ArchTag = ArchTag_;
+
+    static_assert(ArchTag::kMinComputeCapability >= 75);
+    static constexpr bool IsSm90 = ArchTag::kMinComputeCapability >= 90;
+
+    static constexpr uint32_t MaxThreadsPerBlock = kNThreads;
+    static constexpr uint32_t MinBlocksPerMultiprocessor = 2;
+
+    static constexpr int kBlockM = get<0>(TileShape_MK{});
+    static constexpr int kHeadDim = get<1>(TileShape_MK{});
+    static_assert(!IsSm90 || kNThreads % cutlass::NumThreadsPerWarpGroup == 0, "kNThreads must be a multiple of NumThreadsPerWarpGroup");
+    static constexpr int NumdQWarpGgroups = kNThreads / cutlass::NumThreadsPerWarpGroup;
+    using R2SLayoutAtomdQaccum = std::conditional_t<
+        IsSm90,
+        Layout<Shape<Int<cutlass::NumThreadsPerWarpGroup>, Int<NumdQWarpGgroups>>>,
+        Layout<Shape<Int<kNThreads>>>
+    >;
+    using R2STiledCopydQaccum = decltype(make_tiled_copy(Copy_Atom<AutoVectorizingCopyWithAssumedAlignment<128>, ElementAccum>{}, R2SLayoutAtomdQaccum{},
+                                                         Layout<Shape<Int<IsSm90 ? 4 : 1>>>{}));  // Val layout, 1 or 4 vals per read
+    using G2SLayoutAtomdQaccum = Layout<Shape<Int<kNThreads>>>;
+    // UniversalCopy instead of AutoVectorizingCopyWithAssumedAlignment as the latter generates cp.async instructions
+    using G2STiledCopydQaccum = decltype(make_tiled_copy(Copy_Atom<UniversalCopy<uint128_t>, ElementAccum>{}, G2SLayoutAtomdQaccum{},
+                                                         Layout<Shape<_4>>{}));  // Val layout, 4 vals per read
+    // We don't do bound checking for the gmem -> smem load so we just assert here.
+    static_assert(IsSm90 || (kBlockM * kHeadDim) % (kNThreads * 4) == 0);
+    static constexpr int SmemdQaccumSize = size(TileShape_MK{});
+    using SmemLayoutdQaccumFlat = Layout<Shape<Int<SmemdQaccumSize>>>;
+    using SmemLayoutdQaccum = std::conditional_t<
+        IsSm90,
+        Layout<Shape<Int<kBlockM * kHeadDim / NumdQWarpGgroups>, Int<NumdQWarpGgroups>>>,
+        Layout<Shape<Int<kBlockM * kHeadDim>>>
+    >;
+
+    // We can't just use kHeadDim here. E.g. if MMA shape is 64 x 96 but split across 2 WGs,
+    // then setting kBlockKSmem to 32 will cause "Static shape_div failure".
+    // We want to treat it as 64 x 48, so kBlockKSmem should be 16.
+    static constexpr int MmaShapeN = get<1>(typename TiledMma::AtomShape_MNK{});
+    static constexpr int kBlockKSmem = MmaShapeN % 64 == 0 ? 64 : (MmaShapeN % 32 == 0 ? 32 : 16);
+    static constexpr int kSwizzle = kBlockKSmem == 64 ? 3 : (kBlockKSmem == 32 ? 2 : 1);
+    using SmemLayoutAtomdQ =
+        decltype(composition(Swizzle<kSwizzle, 3, 3>{},
+                 Layout<Shape<Int<8>, Int<kBlockKSmem>>,
+                 Stride<Int<kBlockKSmem>, _1>>{}));
+    using SmemLayoutdQ = decltype(tile_to_shape(SmemLayoutAtomdQ{}, TileShape_MK{}));
+    using SmemLayoutdQt =
+        decltype(cute::composition(SmemLayoutdQ{},
+                                   make_layout(make_shape(get<1>(TileShape_MK{}), get<0>(TileShape_MK{})),
+                                               make_stride(Int<get<0>(TileShape_MK{})>{}, _1{}))));
+
+    using SmemCopyAtomdQ = Copy_Atom<
+        std::conditional_t<
+            IsSm90,
+            std::conditional_t<!dQ_swapAB, cute::SM90_U32x4_STSM_N, cute::SM90_U16x8_STSM_T>,
+            AutoVectorizingCopyWithAssumedAlignment<128>
+        >,
+        Element>;
+
+    static constexpr int kGmemElemsPerLoad = sizeof(cute::uint128_t) / sizeof(Element);
+    static_assert(kHeadDim % kGmemElemsPerLoad == 0, "Headdim must be a multiple of kGmemElemsPerLoad");
+    static constexpr int kGmemThreadsPerRow = cutlass::gcd(kHeadDim / kGmemElemsPerLoad, int(MaxThreadsPerBlock));
+    static_assert(MaxThreadsPerBlock % kGmemThreadsPerRow == 0, "MaxThreadsPerBlock must be a multiple of kGmemThreadsPerRow");
+    using GmemLayoutAtom = Layout<Shape <Int<MaxThreadsPerBlock / kGmemThreadsPerRow>, Int<kGmemThreadsPerRow>>,
+                                  Stride<Int<kGmemThreadsPerRow>, _1>>;
+    using GmemTiledCopy = decltype(
+        make_tiled_copy(Copy_Atom<AutoVectorizingCopyWithAssumedAlignment<128>, Element>{},
+                        GmemLayoutAtom{},
+                        Layout<Shape<_1, Int<kGmemElemsPerLoad>>>{}));  // Val layout, 8 or 16 vals per load
+
+    struct SharedStorage : cute::aligned_struct<128> {
+        cute::array_aligned<ElementAccum, cute::cosize_v<SmemLayoutdQaccum>> smem_dqacc;
+        cute::array_aligned<Element, cute::cosize_v<SmemLayoutdQ>> smem_dq;
+        alignas(16) cutlass::arch::ClusterTransactionBarrier barrier_dQaccum;
+    };
+
+    static constexpr int SharedStorageSize = sizeof(SharedStorage);
+
+    using ShapedQ = cute::Shape<int32_t, int32_t, int32_t, int32_t>;   // (seqlen_q, d, head, batch)
+    using StridedQ = cute::Stride<int64_t, _1, int64_t, int64_t>;
+    using ShapedQaccum = cute::Shape<int32_t, int32_t, int32_t>;  // (seqlen_q * d, head, batch)
+    using StridedQaccum = cute::Stride<_1, int64_t, int64_t>;
+
+    // Device side arguments
+    struct Arguments {
+        ElementAccum const* ptr_dQaccum;
+        ShapedQaccum const shape_dQaccum;
+        StridedQaccum const stride_dQaccum;
+        Element* ptr_dQ;
+        ShapedQ const shape_dQ;
+        StridedQ const stride_dQ;
+        float const softmax_scale;
+        int const* cu_seqlens = nullptr;
+        int const* seqused = nullptr;
+    };
+
+    // Kernel entry point API
+    struct Params {
+        ElementAccum const* ptr_dQaccum;
+        ShapedQaccum const shape_dQaccum;
+        StridedQaccum const stride_dQaccum;
+        Element* ptr_dQ;
+        ShapedQ const shape_dQ;
+        StridedQ const stride_dQ;
+        float const softmax_scale;
+        int const* cu_seqlens = nullptr;
+        int const* seqused = nullptr;
+    };
+
+    // Convert to underlying arguments. In this case, a simple copy for the aliased type.
+    static
+    Params
+    to_underlying_arguments(Arguments const& args) {
+        return {
+            args.ptr_dQaccum,
+            args.shape_dQaccum,
+            args.stride_dQaccum,
+            args.ptr_dQ,
+            args.shape_dQ,
+            args.stride_dQ,
+            args.softmax_scale,
+            args.cu_seqlens,
+            args.seqused
+        };
+    }
+
+    CUTLASS_DEVICE
+    void
+    operator()(Params const& params, char* smem_buf) {
+
+        static constexpr int kBlockM = get<0>(TileShape_MK{});
+        SharedStorage& shared_storage = *reinterpret_cast<SharedStorage*>(smem_buf);
+
+        Tensor sdQaccum = make_tensor(make_smem_ptr(shared_storage.smem_dqacc.data()), SmemLayoutdQaccum{});
+        Tensor sdQaccum_flat = make_tensor(make_smem_ptr(shared_storage.smem_dqacc.data()), SmemLayoutdQaccumFlat{});
+        Tensor sdQ = make_tensor(make_smem_ptr(shared_storage.smem_dq.data()), SmemLayoutdQ{});
+        Tensor sdQt = make_tensor(make_smem_ptr(shared_storage.smem_dq.data()), SmemLayoutdQt{});
+
+        int const thread_idx = threadIdx.x;
+        int const m_block = blockIdx.x;
+        int const bidh = blockIdx.y;
+        int const bidb = blockIdx.z;
+
+        flash::SeqlenInfo<true /*Varlen*/, kBlockM> seqlen_info(bidb, size<0>(params.shape_dQ), params.cu_seqlens, params.seqused);
+        bool const is_varlen = params.cu_seqlens;
+        if (is_varlen && m_block * kBlockM >= seqlen_info.seqlen) { return; }
+
+        // Step 1: load dQaccum from gmem to smem
+        Tensor mdQaccum = make_tensor(make_gmem_ptr(reinterpret_cast<ElementAccum const*>(params.ptr_dQaccum)),
+                                      params.shape_dQaccum, params.stride_dQaccum)(_, bidh, !is_varlen ? bidb : 0);
+        Tensor gdQaccum = local_tile(domain_offset(make_coord(seqlen_info.offset_padded * kHeadDim), mdQaccum), Shape<Int<kBlockM * kHeadDim>>{}, make_coord(m_block));  // (M * K)
+        if constexpr (IsSm90) {  // Use BulkCopy
+            static constexpr uint32_t TmaTransactionBytesdQaccum = static_cast<uint32_t>(size(SmemLayoutdQaccumFlat{}) * cute::sizeof_bits_v<ElementAccum> / 8);
+            auto bulk_copy = Copy_Traits<SM90_BULK_COPY_AUTO>{};
+            // if (thread0()) { print(gdQaccum); printf("\n"); print(sdQaccum_flat); printf("\n"); }
+            if (thread_idx == 0) {
+                shared_storage.barrier_dQaccum.init(1 /*numThreads*/);
+                shared_storage.barrier_dQaccum.arrive_and_expect_tx(TmaTransactionBytesdQaccum);
+                copy(bulk_copy.with(*reinterpret_cast<uint64_t*>(&shared_storage.barrier_dQaccum)), gdQaccum, sdQaccum_flat);
+            }
+            __syncthreads();
+            shared_storage.barrier_dQaccum.wait(0);
+        } else {
+            G2STiledCopydQaccum g2s_tiled_copy_dQaccum;
+            auto g2s_thr_copy_dQaccum = g2s_tiled_copy_dQaccum.get_thread_slice(thread_idx);
+            Tensor tdQgdQaccumg2s = g2s_thr_copy_dQaccum.partition_S(gdQaccum);
+            Tensor tdQsdQaccumg2s = g2s_thr_copy_dQaccum.partition_D(sdQaccum);
+            cute::copy(g2s_tiled_copy_dQaccum, tdQgdQaccumg2s, tdQsdQaccumg2s);
+            __syncthreads();
+        }
+
+        // __syncthreads(); if (cute::thread0()) { print_tensor(sdQaccum); }
+
+        // Step 2: Load dQaccum from smem to register, then convert fp32 -> fp16/bf16
+        R2STiledCopydQaccum s2r_tiled_copy_dQaccum;
+        auto s2r_thr_copy_dQaccum = s2r_tiled_copy_dQaccum.get_thread_slice(thread_idx);
+        Tensor tdQsdQaccum = s2r_thr_copy_dQaccum.partition_S(sdQaccum);
+        TiledMma tiled_mma_dQ;
+        Tensor taccdQrdQaccum = partition_fragment_C(tiled_mma_dQ, select<!dQ_swapAB ? 0 : 1, !dQ_swapAB ? 1 : 0>(TileShape_MK{}));
+        // if (blockIdx.x == 0 && blockIdx.y == 0 && threadIdx.x == 1) { print(tiled_mma_dQ); printf("\n"); }
+        // if (blockIdx.x == 0 && blockIdx.y == 0 && threadIdx.x == 1) { print(tdQsdQaccum); }
+        // if (blockIdx.x == 0 && blockIdx.y == 0 && threadIdx.x == 1) { print(taccdQrdQaccum); }
+        CUTE_STATIC_ASSERT_V(size(taccdQrdQaccum) == size(tdQsdQaccum));
+        Tensor tdQrdQaccum = s2r_thr_copy_dQaccum.retile_D(taccdQrdQaccum);
+        cute::copy(s2r_tiled_copy_dQaccum, tdQsdQaccum, tdQrdQaccum);
+        #pragma unroll
+        for (int i = 0; i < size(taccdQrdQaccum); ++i) { taccdQrdQaccum(i) *= params.softmax_scale; }
+        // Convert tdQrdQ from fp32 to fp16
+        Tensor rdQ = make_tensor_like<Element>(taccdQrdQaccum);
+        flash::convert_type_out(taccdQrdQaccum, rdQ);
+
+        // Step 3: Copy dQ from register to smem
+        auto smem_tiled_copy_dQ = make_tiled_copy_C(SmemCopyAtomdQ{}, tiled_mma_dQ);
+        auto smem_thr_copy_dQ = smem_tiled_copy_dQ.get_thread_slice(thread_idx);
+        Tensor taccdQrdQ = smem_thr_copy_dQ.retile_S(rdQ);  // ((Atom,AtomNum), MMA_N, MMA_N)
+        // if (cute::thread0()) { print(smem_tiled_copy_dQ); }
+        // if (cute::thread0()) { print(smem_thr_copy_dQ); }
+        // if (cute::thread0()) { print(sdQ); }
+        Tensor taccdQsdQ = smem_thr_copy_dQ.partition_D(cute::conditional_return<!dQ_swapAB>(sdQ, sdQt));  // ((Atom,AtomNum),PIPE_M,PIPE_N)
+        cute::copy(smem_tiled_copy_dQ, taccdQrdQ, taccdQsdQ);
+        __syncthreads();
+
+        // Step 4: Copy dQ from smem to register to prepare for coalesced write to gmem
+        Tensor mdQ = make_tensor(make_gmem_ptr(params.ptr_dQ), params.shape_dQ, params.stride_dQ)(_, _, bidh, !is_varlen ? bidb : 0);
+        Tensor gdQ = local_tile(domain_offset(make_coord(seqlen_info.offset, _0{}), mdQ), TileShape_MK{}, make_coord(m_block, _0{}));  // (M, K)
+        GmemTiledCopy gmem_tiled_copy_dQ;
+        auto gmem_thr_copy_dQ = gmem_tiled_copy_dQ.get_thread_slice(thread_idx);
+        Tensor tdQsdQ = gmem_thr_copy_dQ.partition_S(sdQ);    // ((Atom,AtomNum),ATOM_M,ATOM_N)
+        Tensor tdQgdQ = gmem_thr_copy_dQ.partition_D(gdQ);
+
+        Tensor tdQrdQ = make_fragment_like(tdQsdQ);
+        Tensor tdQcdQ = gmem_thr_copy_dQ.partition_D(cute::make_identity_tensor(TileShape_MK{}));
+        Tensor tdQpdQ = make_tensor<bool>(make_shape(size<2>(tdQgdQ)));
+        #pragma unroll
+        for (int k = 0; k < size(tdQpdQ); ++k) { tdQpdQ(k) = get<1>(tdQcdQ(_0{}, _0{}, k)) < get<1>(params.shape_dQ); }
+        // Need to check OOB when reading from smem if kBlockM isn't evenly tiled
+        static constexpr bool EvenM = kBlockM % CUTE_STATIC_V(size<0>(GmemLayoutAtom{})) == 0;
+        flash::copy</*Is_even_MN=*/EvenM, /*Is_even_K=*/true, /*Clear_OOB_MN=*/false>(
+            gmem_tiled_copy_dQ, tdQsdQ, tdQrdQ, tdQcdQ, tdQpdQ, kBlockM);
+
+        // Step 5: Copy dQ from register to gmem
+        // Clear_OOB_K must be false since we don't want to write zeros to gmem
+        flash::copy</*Is_even_MN=*/false, /*Is_even_K=*/false, /*Clear_OOB_MN=*/false, /*Clear_OOB_K=*/false>(
+            gmem_tiled_copy_dQ, tdQrdQ, tdQgdQ, tdQcdQ, tdQpdQ, std::min(seqlen_info.seqlen - m_block * kBlockM, kBlockM)
+        );
+    }
+
+};
+
+} // namespace flash

Code/Baselines/flash-attention/hopper/padding.py

@@ -0,0 +1,53 @@
+# Adapted from https://github.com/mlcommons/training_results_v1.1/blob/main/NVIDIA/benchmarks/bert/implementations/pytorch/padding.py
+
+import torch
+import torch.nn.functional as F
+from einops import rearrange
+
+
+def unpad_input(hidden_states, attention_mask, unused_mask=None):
+    """
+    Arguments:
+        hidden_states: (batch, seqlen, ...)
+        attention_mask: (batch, seqlen), bool / int, 1 means valid and 0 means not valid.
+        unused_mask: (batch, seqlen), bool / int, 1 means the element is allocated but unused.
+    Return:
+        hidden_states: (total_nnz, ...), where total_nnz = number of tokens selected in attention_mask + unused_mask.
+        indices: (total_nnz), the indices of masked tokens from the flattened input sequence.
+        cu_seqlens: (batch + 1), the cumulative sequence lengths, used to index into hidden_states.
+        max_seqlen_in_batch: int
+        seqused: (batch), returns the number of tokens selected in attention_mask + unused_mask.
+    """
+    all_masks = (attention_mask + unused_mask) if unused_mask is not None else attention_mask
+    seqlens_in_batch = all_masks.sum(dim=-1, dtype=torch.int32)
+    used_seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
+    indices = torch.nonzero(all_masks.flatten(), as_tuple=False).flatten()
+    max_seqlen_in_batch = seqlens_in_batch.max().item()
+    cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))
+    # TD [2022-03-04] We don't want to index with a bool mask, because Pytorch will expand the
+    # bool mask, then call nonzero to get the indices, then index with those. The indices is @dim
+    # times larger than it needs to be, wasting memory. It's faster and more memory-efficient to
+    # index with integer indices.
+    return (
+        rearrange(hidden_states, "b s ... -> (b s) ...")[indices],
+        indices,
+        cu_seqlens,
+        max_seqlen_in_batch,
+        used_seqlens_in_batch,
+    )
+
+
+def pad_input(hidden_states, indices, batch, seqlen):
+    """
+    Arguments:
+        hidden_states: (total_nnz, ...), where total_nnz = number of tokens in selected in attention_mask.
+        indices: (total_nnz), the indices that represent the non-masked tokens of the original padded input sequence.
+        batch: int, batch size for the padded sequence.
+        seqlen: int, maximum sequence length for the padded sequence.
+    Return:
+        hidden_states: (batch, seqlen, ...)
+    """
+    dim = hidden_states.shape[1:]
+    output = torch.zeros((batch * seqlen), *dim, device=hidden_states.device, dtype=hidden_states.dtype)
+    output[indices] = hidden_states
+    return rearrange(output, "(b s) ... -> b s ...", b=batch)

Code/Baselines/flash-attention/hopper/static_switch.h

@@ -0,0 +1,181 @@
+// Inspired by
+// https://github.com/NVIDIA/DALI/blob/main/include/dali/core/static_switch.h
+// and https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/Dispatch.h
+
+#pragma once
+
+/// @param COND       - a boolean expression to switch by
+/// @param CONST_NAME - a name given for the constexpr bool variable.
+/// @param ...       - code to execute for true and false
+///
+/// Usage:
+/// ```
+/// BOOL_SWITCH(flag, BoolConst, [&] {
+///     some_function<BoolConst>(...);
+/// });
+/// ```
+//
+
+#define BOOL_SWITCH(COND, CONST_NAME, ...)                                                       \
+  [&] {                                                                                          \
+    if (COND) {                                                                                  \
+      constexpr static bool CONST_NAME = true;                                                   \
+      return __VA_ARGS__();                                                                      \
+    } else {                                                                                     \
+      constexpr static bool CONST_NAME = false;                                                  \
+      return __VA_ARGS__();                                                                      \
+    }                                                                                            \
+  }()
+
+#ifdef FLASHATTENTION_DISABLE_LOCAL
+  #define CAUSAL_LOCAL_SWITCH(CAUSAL_COND, LOCAL_COND, CAUSAL_CONST_NAME, LOCAL_CONST_NAME, ...) \
+    [&] {                                                                                        \
+      constexpr static bool LOCAL_CONST_NAME = false;                                            \
+      if (CAUSAL_COND) {                                                                         \
+        constexpr static bool CAUSAL_CONST_NAME = true;                                          \
+        return __VA_ARGS__();                                                                    \
+      } else {                                                                                   \
+        constexpr static bool CAUSAL_CONST_NAME = false;                                         \
+        return __VA_ARGS__();                                                                    \
+      }                                                                                          \
+    }()
+#else
+  #define CAUSAL_LOCAL_SWITCH(CAUSAL_COND, LOCAL_COND, CAUSAL_CONST_NAME, LOCAL_CONST_NAME, ...) \
+    [&] {                                                                                        \
+      if (CAUSAL_COND) {                                                                         \
+        constexpr static bool CAUSAL_CONST_NAME = true;                                          \
+        constexpr static bool LOCAL_CONST_NAME = false;                                          \
+        return __VA_ARGS__();                                                                    \
+      } else if (LOCAL_COND) {                                                                   \
+        constexpr static bool CAUSAL_CONST_NAME = false;                                         \
+        constexpr static bool LOCAL_CONST_NAME = true;                                           \
+        return __VA_ARGS__();                                                                    \
+      } else {                                                                                   \
+        constexpr static bool CAUSAL_CONST_NAME = false;                                         \
+        constexpr static bool LOCAL_CONST_NAME = false;                                          \
+        return __VA_ARGS__();                                                                    \
+      }                                                                                          \
+    }()
+#endif
+
+#ifdef FLASHATTENTION_DISABLE_SOFTCAP
+  #define SOFTCAP_SWITCH(COND, CONST_NAME, ...)                                                  \
+  [&] {                                                                                          \
+    constexpr static bool CONST_NAME = false;                                                    \
+    return __VA_ARGS__();                                                                        \
+  }()
+#else
+  #define SOFTCAP_SWITCH BOOL_SWITCH
+#endif
+
+#ifdef FLASHATTENTION_DISABLE_PAGEDKV
+  #define PAGEDKV_SWITCH(COND, CONST_NAME, ...)                                                  \
+  [&] {                                                                                          \
+    constexpr static bool CONST_NAME = false;                                                    \
+    return __VA_ARGS__();                                                                        \
+  }()
+#else
+  #define PAGEDKV_SWITCH BOOL_SWITCH
+#endif
+
+#ifdef FLASHATTENTION_DISABLE_SPLIT
+  #define SPLIT_SWITCH(COND, CONST_NAME, ...)                                                    \
+  [&] {                                                                                          \
+    constexpr static bool CONST_NAME = false;                                                    \
+    return __VA_ARGS__();                                                                        \
+  }()
+#else
+  #define SPLIT_SWITCH BOOL_SWITCH
+#endif
+
+#ifdef FLASHATTENTION_DISABLE_APPENDKV
+  #define APPENDKV_SWITCH(COND, CONST_NAME, ...)                                                 \
+  [&] {                                                                                          \
+    constexpr static bool CONST_NAME = false;                                                    \
+    return __VA_ARGS__();                                                                        \
+  }()
+#else
+  #define APPENDKV_SWITCH BOOL_SWITCH
+#endif
+
+#ifdef FLASHATTENTION_DISABLE_PACKGQA
+  #define PACKGQA_SWITCH(COND, CONST_NAME, ...)                                                  \
+  [&] {                                                                                          \
+    constexpr static bool CONST_NAME = false;                                                    \
+    return __VA_ARGS__();                                                                        \
+  }()
+#else
+  #define PACKGQA_SWITCH BOOL_SWITCH
+#endif
+
+#ifdef FLASHATTENTION_DISABLE_VARLEN
+  #define VARLEN_SWITCH(COND, CONST_NAME, ...)                                                   \
+  [&] {                                                                                          \
+    constexpr static bool CONST_NAME = false;                                                    \
+    return __VA_ARGS__();                                                                        \
+  }()
+#else
+  #define VARLEN_SWITCH BOOL_SWITCH
+#endif
+
+#ifdef FLASHATTENTION_DISABLE_CLUSTER
+  #define CLUSTER_SWITCH(COND, CONST_NAME, ...)                                                  \
+  [&] {                                                                                          \
+    constexpr static bool CONST_NAME = false;                                                    \
+    return __VA_ARGS__();                                                                        \
+  }()
+#else
+  #define CLUSTER_SWITCH BOOL_SWITCH
+#endif
+
+#ifdef FLASHATTENTION_DISABLE_SM8x
+  #define ARCH_SWITCH(ARCH, ARCH_NAME, ...)                                                      \
+  [&] {                                                                                          \
+    constexpr static int ARCH_NAME = 90;                                                         \
+    return __VA_ARGS__();                                                                        \
+  }()
+#else
+  #define ARCH_SWITCH(ARCH, ARCH_NAME, ...)                                                      \
+  [&] {                                                                                          \
+    if (ARCH == 86 || ARCH == 89) {                                                              \
+      constexpr static int ARCH_NAME = 86;                                                       \
+      return __VA_ARGS__();                                                                      \
+    } else if (ARCH < 90) {                                                                      \
+      constexpr static int ARCH_NAME = 80;                                                       \
+      return __VA_ARGS__();                                                                      \
+    } else {                                                                                     \
+      constexpr static int ARCH_NAME = 90;                                                       \
+      return __VA_ARGS__();                                                                      \
+    }                                                                                            \
+  }()
+#endif
+
+#ifndef FLASHATTENTION_ENABLE_VCOLMAJOR
+  #define VCOLMAJOR_SWITCH(COND, CONST_NAME, ...)                                                \
+  [&] {                                                                                          \
+    constexpr static bool CONST_NAME = false;                                                    \
+    return __VA_ARGS__();                                                                        \
+  }()
+#else
+  #define VCOLMAJOR_SWITCH BOOL_SWITCH
+#endif
+
+#define HEADDIM_SWITCH(HEADDIM, ...)                                                             \
+  [&] {                                                                                          \
+    if (HEADDIM == 64) {                                                                         \
+      constexpr static int kHeadSize = 64;                                                       \
+      return __VA_ARGS__();                                                                      \
+    } else if (HEADDIM == 96) {                                                                  \
+      constexpr static int kHeadSize = 96;                                                       \
+      return __VA_ARGS__();                                                                      \
+    } else if (HEADDIM == 128) {                                                                 \
+      constexpr static int kHeadSize = 128;                                                      \
+      return __VA_ARGS__();                                                                      \
+    } else if (HEADDIM == 96) {                                                                  \
+      constexpr static int kHeadSize = 96;                                                       \
+      return __VA_ARGS__();                                                                      \
+    } else if (HEADDIM == 256) {                                                                 \
+      constexpr static int kHeadSize = 256;                                                      \
+      return __VA_ARGS__();                                                                      \
+    }                                                                                            \
+  }()

Code/Baselines/flash-attention/hopper/test_flash_attn.py

@@ -0,0 +1,1171 @@
+import os
+import math
+import itertools
+
+import pytest
+import torch
+import torch.nn.functional as F
+from torch._C import parse_schema
+
+from einops import rearrange, repeat
+try:
+    from flash_attn.layers.rotary import apply_rotary_emb
+except ImportError:
+    apply_rotary_emb = None
+
+from padding import pad_input, unpad_input
+from test_util import (
+    attention_ref,
+    generate_qkv,
+    generate_random_padding_mask,
+)
+
+from flash_attn_interface import flash_attn_func, flash_attn_varlen_func, flash_attn_combine
+from flash_attn_interface import flash_attn_with_kvcache, get_scheduler_metadata
+
+
+DISABLE_BACKWARD = os.getenv("FLASH_ATTENTION_DISABLE_BACKWARD", "FALSE") == "TRUE"
+DISABLE_SPLIT = os.getenv("FLASH_ATTENTION_DISABLE_SPLIT", "FALSE") == "TRUE"
+DISABLE_PAGEDKV = os.getenv("FLASH_ATTENTION_DISABLE_PAGEDKV", "FALSE") == "TRUE"
+DISABLE_APPENDKV = os.getenv("FLASH_ATTENTION_DISABLE_APPENDKV", "FALSE") == "TRUE"
+DISABLE_LOCAL = os.getenv("FLASH_ATTENTION_DISABLE_LOCAL", "FALSE") == "TRUE"
+DISABLE_SOFTCAP = os.getenv("FLASH_ATTENTION_DISABLE_SOFTCAP", "FALSE") == "TRUE"
+DISABLE_PACKGQA = os.getenv("FLASH_ATTENTION_DISABLE_PACKGQA", "FALSE") == "TRUE"
+DISABLE_FP16 = os.getenv("FLASH_ATTENTION_DISABLE_FP16", "FALSE") == "TRUE"
+DISABLE_FP8 = os.getenv("FLASH_ATTENTION_DISABLE_FP8", "FALSE") == "TRUE" or torch.cuda.get_device_capability("cuda")[0] < 9
+DISABLE_HDIM64 = os.getenv("FLASH_ATTENTION_DISABLE_HDIM64", "FALSE") == "TRUE"
+DISABLE_HDIM96 = os.getenv("FLASH_ATTENTION_DISABLE_HDIM96", "FALSE") == "TRUE"
+DISABLE_HDIM128 = os.getenv("FLASH_ATTENTION_DISABLE_HDIM128", "FALSE") == "TRUE"
+DISABLE_HDIM192 = os.getenv("FLASH_ATTENTION_DISABLE_HDIM192", "FALSE") == "TRUE"
+DISABLE_HDIM256 = os.getenv("FLASH_ATTENTION_DISABLE_HDIM256", "FALSE") == "TRUE"
+
+COMPILED_HDIMS = (
+    []
+    + ([64] if not DISABLE_HDIM64 else [])
+    + ([96] if not DISABLE_HDIM96 else [])
+    + ([128] if not DISABLE_HDIM128 else [])
+    + ([192] if not DISABLE_HDIM192 else [])
+    + ([256] if not DISABLE_HDIM256 else [])
+)
+
+
+# @pytest.mark.parametrize("dtype", [torch.float16, torch.bfloat16, torch.float8_e4m3fn])
+@pytest.mark.parametrize("dtype", [torch.bfloat16] + ([torch.float16] if not DISABLE_FP16 else []) + ([torch.float8_e4m3fn] if not DISABLE_FP8 else []))
+# @pytest.mark.parametrize("dtype", [torch.bfloat16])
+# @pytest.mark.parametrize("dtype", [torch.float8_e4m3fn])
+@pytest.mark.parametrize("mha_type", ["mha", "mqa", "gqa"])
+# @pytest.mark.parametrize("mha_type", ["mha"])
+# @pytest.mark.parametrize("has_qv", [False, True])
+@pytest.mark.parametrize("has_qv", [False])
+# @pytest.mark.parametrize("deterministic", [False, True])
+@pytest.mark.parametrize("deterministic", [False])
+@pytest.mark.parametrize("softcap", [0.0] + ([15.0] if not DISABLE_SOFTCAP else []))
+# @pytest.mark.parametrize("softcap", [0.0])
+@pytest.mark.parametrize("local", [False] + ([True] if not DISABLE_LOCAL else []))
+# @pytest.mark.parametrize("local", [False])
+@pytest.mark.parametrize("causal", [False, True])
+# @pytest.mark.parametrize("causal", [True])
+# @pytest.mark.parametrize("V_colmajor", [False, True])
+@pytest.mark.parametrize("V_colmajor", [False])
+# @pytest.mark.parametrize("d", [32, 64, 96, 128, 160, 192, 224, 256])
+# @pytest.mark.parametrize('d', [32, 40, 64, 80, 96, 128, 160, 192, 256])
+# @pytest.mark.parametrize('d', [32, 64, 96, 128, 160, 192])
+# @pytest.mark.parametrize('d', [56, 80])
+# @pytest.mark.parametrize("d", [64, 128, 256])
+# @pytest.mark.parametrize('d', [32, 40, 64, 80, 96, 128])
+# @pytest.mark.parametrize("d", [64, 96, 128, 192])
+@pytest.mark.parametrize("d", COMPILED_HDIMS)
+# @pytest.mark.parametrize("d", [128])
+@pytest.mark.parametrize(
+    "seqlen_q,seqlen_k",
+    [
+        (1, 1),
+        (64, 128),
+        (128, 192),
+        (256, 256),
+        (239, 1),
+        (799, 3),
+        (113, 203),
+        (113, 128),
+        (128, 217),
+        (113, 211),
+        (108, 256),
+        (256, 512),
+        (384, 256),
+        (640, 128),
+        (512, 256),
+        (1024, 1024),
+        (1023, 1024),
+        (1024, 1023),
+        (4096, 4096),
+        (4224, 4224),
+    ],
+)
+# @pytest.mark.parametrize('seqlen_q,seqlen_k', [(128, 128)])
+def test_flash_attn_output(
+        seqlen_q, seqlen_k, d, causal, local, softcap, V_colmajor, deterministic, has_qv, mha_type, dtype
+):
+    if V_colmajor and (seqlen_k % 16 != 0 or dtype != torch.float8_e4m3fn):
+        pytest.skip("V_colmajor requires seqlen_k to be a multiple of 16 and dtype to be float8_e4m3fn")
+    device = "cuda"
+    # set seed
+    torch.random.manual_seed(0)
+    # batch_size = 40
+    # nheads = 16
+    batch_size = 9 if seqlen_k <= 2048 else 2
+    # batch_size = 1
+    nheads = 6
+    # nheads = 1
+    nheads_kv = nheads if mha_type == "mha" else (2 if mha_type == "gqa" else 1)
+    dtype_ref = torch.bfloat16 if dtype == torch.float8_e4m3fn else dtype
+    dv_vals = [128, d] if d > 128 and d <= 192 else ([256, 512, d] if d <= 64 else [d])
+    if dtype == torch.float8_e4m3fn:
+        dv_vals = [d]
+    attention_chunk_vals = [torch.randint(1, seqlen_k * 2, (1,)).item(), 0] if not DISABLE_LOCAL else [0]
+    for dv, attention_chunk in itertools.product(dv_vals, attention_chunk_vals):
+        q_ref = torch.randn(batch_size, seqlen_q, nheads, d, device=device, dtype=dtype_ref)
+        if softcap > 0.0:
+            # Ensure the values of qk are at least within softcap range.
+            q_ref = (q_ref * softcap / 4)
+        q_ref = q_ref.to(dtype).to(dtype_ref).requires_grad_()
+        k_ref = torch.randn(batch_size, seqlen_k, nheads_kv, d, device=device, dtype=dtype_ref).to(dtype).to(dtype_ref).requires_grad_()
+        v_ref = torch.randn(batch_size, seqlen_k, nheads_kv, dv, device=device, dtype=dtype_ref).to(dtype).to(dtype_ref).requires_grad_()
+        if has_qv:
+            qv_ref = torch.randn(batch_size, seqlen_q, nheads, dv, device=device, dtype=dtype_ref).to(dtype).to(dtype_ref)
+        else:
+            qv_ref = None
+        # Put window_size after QKV randn so that window_size changes from test to test
+        window_size = (-1, -1) if not local else torch.randint(0, seqlen_k, (2,)).tolist()
+        # window_size = (-1, -1) if not local else (16, 0)
+        if dtype == torch.float8_e4m3fn:
+            q_descale, k_descale, v_descale = [torch.rand(batch_size, nheads_kv, device=device, dtype=torch.float32) * 2 for _ in range(3)]
+        else:
+            q_descale, k_descale, v_descale = None, None, None
+        q, k, v = [x.detach().to(dtype).requires_grad_() for x in (q_ref, k_ref, v_ref)]
+        qv = qv_ref.detach().to(dtype).requires_grad_() if has_qv else None
+        if V_colmajor:
+            v = rearrange(rearrange(v.detach(), "b s h d -> b h d s").contiguous(), "b h d s -> b s h d").requires_grad_()
+        out_ref, attn_ref = attention_ref(
+            q_ref,
+            k_ref,
+            v_ref,
+            None,
+            None,
+            causal=causal,
+            qv=qv_ref,
+            q_descale=q_descale, k_descale=k_descale, v_descale=v_descale,
+            window_size=window_size,
+            attention_chunk=attention_chunk,
+            softcap=softcap
+        )
+        out_pt, attn_pt = attention_ref(
+            q_ref,
+            k_ref,
+            v_ref,
+            None,
+            None,
+            causal=causal,
+            qv=qv_ref,
+            q_descale=q_descale, k_descale=k_descale, v_descale=v_descale,
+            window_size=window_size,
+            attention_chunk=attention_chunk,
+            softcap=softcap,
+            upcast=False,
+            reorder_ops=True,
+            intermediate_dtype=dtype if dtype == torch.float8_e4m3fn else None,
+        )
+
+        # qk = torch.einsum('bshd,bthd->bhst', q_ref, k_ref).float()
+        # if qv is not None:
+        #     qk += torch.einsum('bshd,bthd->bhst', qv_ref, v_ref).float()
+        # m = qk.amax(-1, keepdim=True)
+        # s_tmp = torch.exp((qk - m) / math.sqrt(d))
+        # exp_sum = s_tmp.sum(-1)
+        # qk = torch.einsum('bthd,bshd->bhts', q_ref.float() / math.sqrt(d), k_ref.float())
+        # lse_ref = torch.logsumexp(qk, dim=-1)
+
+        # Numerical error if we just do any arithmetic on out_ref
+        fwd_atol = 2 * (out_ref + 0.3 - 0.3 - out_ref).abs().max().item()
+        rtol = 2 if softcap == 0.0 else 3
+
+        print(f"Pytorch max diff: {(out_pt - out_ref).abs().max().item()}")
+        print(f"Pytorch mean diff: {(out_pt - out_ref).abs().mean().item()}")
+        pack_gqa_vals = [False, True] if not DISABLE_PACKGQA else [False]
+        num_splits_vals = [1, 3] if not DISABLE_SPLIT else [1]
+        for pack_gqa, num_splits in itertools.product(pack_gqa_vals, num_splits_vals):
+            out = flash_attn_func(
+                q,
+                k,
+                v,
+                causal=causal,
+                qv=qv,
+                q_descale=q_descale, k_descale=k_descale, v_descale=v_descale,
+                window_size=window_size,
+                attention_chunk=attention_chunk,
+                softcap=softcap,
+                pack_gqa=pack_gqa,
+                num_splits=num_splits
+            )
+            print(f"Output max diff: {(out - out_ref).abs().max().item()}")
+            print(f"Output mean diff: {(out - out_ref).abs().mean().item()}")
+            # if not causal:
+            #     print(f"LSE max diff: {(lse - lse_ref).abs().max().item()}")
+            # breakpoint()
+
+            # Check that FlashAttention's numerical error is at most twice the numerical error
+            # of a Pytorch implementation.
+            assert (out - out_ref).abs().max().item() <= rtol * (out_pt - out_ref).abs().max().item() + fwd_atol
+
+        if (
+            not DISABLE_BACKWARD 
+            and dtype != torch.float8_e4m3fn 
+            and not V_colmajor 
+            and not has_qv
+            and not dv > 256
+            and not attention_chunk != 0
+        ):
+            g = torch.randn_like(out)
+            do_o = ((g.float() * out.float()).sum(-1)).transpose(1, 2)
+            # import flash_attn_3_cuda
+            # dq, dk, dv, softmax_d, dq_accum, dk_accum, dv_accum = flash_attn_3_cuda.bwd(
+            #     g,
+            #     q,
+            #     k,
+            #     v,
+            #     out,
+            #     lse,
+            #     None,
+            #     None,
+            #     None,
+            #     d ** (-0.5),
+            #     causal,
+            #     window_size[0], window_size[1],
+            #     softcap,
+            #     deterministic,
+            #     0,  # sm_margin
+            # )
+            dq, dk, dv = torch.autograd.grad(out, (q, k, v), g)
+            # print(f"dO_O max diff: {(softmax_d - do_o).abs().max().item()}")
+            # assert (softmax_d - do_o).abs().max().item() <= 1e-5
+            # assert dq_accum.abs().max().item() == 0.0
+
+            # dS = torch.einsum('bthd,bshd->bhts', g.float(), v.float())
+            # P = torch.softmax(qk, -1)
+            # dP = P * (dS - do_o.transpose(1, 2).unsqueeze(1))
+            # dQ = torch.einsum('bhts,bshd->bthd', dP, k.float())
+            # dV = torch.einsum('bhts,bthd->bshd', P, g.float())
+            # dK = torch.einsum('bhts,bthd->bshd', dP, q.float())
+
+            # dq, dk, dv = torch.autograd.grad(out, (q, k, v), g)
+            dq_ref, dk_ref, dv_ref = torch.autograd.grad(out_ref, (q_ref, k_ref, v_ref), g)
+            dq_pt, dk_pt, dv_pt = torch.autograd.grad(out_pt, (q_ref, k_ref, v_ref), g)
+            print(f"dQ max diff: {(dq - dq_ref).abs().max().item()}")
+            print(f"dK max diff: {(dk - dk_ref).abs().max().item()}")
+            print(f"dV max diff: {(dv - dv_ref).abs().max().item()}")
+            print(f"dQ mean diff: {(dq - dq_ref).abs().mean().item()}")
+            print(f"dK mean diff: {(dk - dk_ref).abs().mean().item()}")
+            print(f"dV mean diff: {(dv - dv_ref).abs().mean().item()}")
+            print(f"dQ Pytorch max diff: {(dq_pt - dq_ref).abs().max().item()}")
+            print(f"dK Pytorch max diff: {(dk_pt - dk_ref).abs().max().item()}")
+            print(f"dV Pytorch max diff: {(dv_pt - dv_ref).abs().max().item()}")
+            print(f"dQ Pytorch mean diff: {(dq_pt - dq_ref).abs().mean().item()}")
+            print(f"dK Pytorch mean diff: {(dk_pt - dk_ref).abs().mean().item()}")
+            print(f"dV Pytorch mean diff: {(dv_pt - dv_ref).abs().mean().item()}")
+            # breakpoint()
+            dq_atol = 2 * (dq_ref + 0.3 - 0.3 - dq_ref).abs().max().item() + (0 if softcap == 0 else 3e-4)
+            assert (dq - dq_ref).abs().max().item() <= rtol * (dq_pt - dq_ref).abs().max().item() + dq_atol
+            dk_atol = 2 * (dk_ref + 0.3 - 0.3 - dk_ref).abs().max().item() + (0 if softcap == 0 else 3e-4)
+            assert (dk - dk_ref).abs().max().item() <= rtol * (dk_pt - dk_ref).abs().max().item() + dk_atol
+            dv_atol = 2 * (dv_ref + 0.3 - 0.3 - dv_ref).abs().max().item() + (0 if softcap == 0 else 3e-4)
+            assert (dv - dv_ref).abs().max().item() <= rtol * (dv_pt - dv_ref).abs().max().item() + dv_atol
+
+
+# @pytest.mark.parametrize("dtype", [torch.float16, torch.bfloat16, torch.float8_e4m3fn])
+@pytest.mark.parametrize("dtype", [torch.bfloat16] + ([torch.float16] if not DISABLE_FP16 else []) + ([torch.float8_e4m3fn] if not DISABLE_FP8 else []))
+# @pytest.mark.parametrize("dtype", [torch.bfloat16])
+# @pytest.mark.parametrize("dtype", [torch.float8_e4m3fn])
+@pytest.mark.parametrize("mha_type", ["mha", "mqa", "gqa"])
+# @pytest.mark.parametrize("mha_type", ["mha"])
+# @pytest.mark.parametrize("has_qv", [False, True])
+@pytest.mark.parametrize("has_qv", [False])
+# @pytest.mark.parametrize("deterministic", [False, True])
+@pytest.mark.parametrize("deterministic", [False])
+@pytest.mark.parametrize("softcap", [0.0] + ([15.0] if not DISABLE_SOFTCAP else []))
+# @pytest.mark.parametrize("softcap", [0.0])
+@pytest.mark.parametrize("local", [False] + ([True] if not DISABLE_LOCAL else []))
+# @pytest.mark.parametrize("local", [False])
+@pytest.mark.parametrize("causal", [False, True])
+# @pytest.mark.parametrize("causal", [False])
+@pytest.mark.parametrize("add_unused_qkv", [False, True])
+# @pytest.mark.parametrize("add_unused_qkv", [True])
+# @pytest.mark.parametrize("d", [32, 64, 96, 128, 160, 192, 224, 256])
+# @pytest.mark.parametrize('d', [32, 40, 64, 80, 96, 128, 160, 192, 256])
+# @pytest.mark.parametrize('d', [32, 64, 96, 128, 160, 192])
+# @pytest.mark.parametrize('d', [56, 80])
+# @pytest.mark.parametrize('d', [32, 40, 64, 80, 96, 128])
+# @pytest.mark.parametrize("d", [64, 96, 128])
+@pytest.mark.parametrize("d", COMPILED_HDIMS)
+# @pytest.mark.parametrize("d", [128])
+@pytest.mark.parametrize(
+    "seqlen_q,seqlen_k",
+    [
+        (1, 1),
+        (1, 3),
+        (2, 1),
+        (511, 1),
+        (3, 513),
+        (64, 128),
+        (128, 128),
+        (256, 256),
+        (113, 203),
+        (128, 217),
+        (113, 211),
+        (108, 256),
+        (256, 512),
+        (307, 256),
+        (640, 128),
+        (512, 256),
+        (1024, 1024),
+        (1023, 1024),
+        (1024, 1023),
+        (2048, 2048),
+    ],
+)
+def test_flash_attn_varlen_output(
+        seqlen_q, seqlen_k, d, add_unused_qkv, causal, local, softcap, deterministic, has_qv, mha_type, dtype
+):
+    device = "cuda"
+    # set seed
+    torch.random.manual_seed(seqlen_q + seqlen_k + d + int(causal) * 2 + int(local))
+    # batch_size = 40
+    # nheads = 16
+    batch_size = 9 if seqlen_q <= 2048 else 2
+    nheads = 6
+    # batch_size = 2
+    # nheads = 1
+    nheads_kv = nheads if mha_type == "mha" else (2 if mha_type == "gqa" else 1)
+    dtype_ref = torch.bfloat16 if dtype == torch.float8_e4m3fn else dtype
+    dv_vals = [128, d] if d > 128 and d <= 192 else ([256, 512, d] if d <= 64 else [d])
+    if dtype == torch.float8_e4m3fn:
+        dv_vals = [d]
+    attention_chunk_vals = [torch.randint(1, seqlen_k * 2, (1,)).item(), 0] if seqlen_q <= seqlen_k and not DISABLE_LOCAL else [0]
+    for dv, attention_chunk in itertools.product(dv_vals, attention_chunk_vals):
+        q_ref = torch.randn(batch_size, seqlen_q, nheads, d, device=device, dtype=dtype_ref)
+        if softcap > 0.0:
+            # Ensure the values of qk are at least within softcap range.
+            q_ref = (q_ref * softcap / 4).detach().requires_grad_()
+        q_ref = q_ref.to(dtype).to(dtype_ref).requires_grad_()
+        k_ref = torch.randn(batch_size, seqlen_k, nheads_kv, d, device=device, dtype=dtype_ref).to(dtype).to(dtype_ref).requires_grad_()
+        v_ref = torch.randn(batch_size, seqlen_k, nheads_kv, dv, device=device, dtype=dtype_ref).to(dtype).to(dtype_ref).requires_grad_()
+        if has_qv:
+            qv_ref = torch.randn(batch_size, seqlen_q, nheads, dv, device=device, dtype=dtype_ref).to(dtype).to(dtype_ref)
+        else:
+            qv_ref = None
+        # Put window_size after QKV randn so that window_size changes from test to test
+        window_size = (-1, -1) if not local else torch.randint(0, seqlen_k, (2,))
+        if dtype == torch.float8_e4m3fn:
+            q_descale, k_descale, v_descale = [torch.rand(batch_size, nheads_kv, device=device, dtype=torch.float32) * 2 for _ in range(3)]
+        else:
+            q_descale, k_descale, v_descale = None, None, None
+        q, k, v = [x.detach().requires_grad_() for x in (q_ref, k_ref, v_ref)]
+        qv = qv_ref.detach() if has_qv else None
+        query_padding_mask = generate_random_padding_mask(
+            seqlen_q, batch_size, device, mode="random", zero_lengths=False
+        )
+        key_padding_mask = generate_random_padding_mask(
+            seqlen_k, batch_size, device, mode="random", zero_lengths=True
+        )
+
+        def _gen_unused_masks(padding_mask, add_unused, max_seq_len, bs, device):
+            if add_unused:
+                another_mask = generate_random_padding_mask(max_seq_len, bs, device)
+                attn_mask = torch.logical_and(padding_mask, another_mask)
+                unused_mask = torch.logical_xor(
+                    torch.logical_or(padding_mask, another_mask), attn_mask
+                )
+            else:
+                attn_mask = padding_mask
+                unused_mask = None
+            return attn_mask, unused_mask
+
+        query_padding_mask, query_unused_mask = _gen_unused_masks(
+            query_padding_mask, add_unused_qkv, seqlen_q, batch_size, q.device
+        )
+        key_padding_mask, key_unused_mask = _gen_unused_masks(
+            key_padding_mask, add_unused_qkv, seqlen_k, batch_size, k.device
+        )
+
+        (
+            q_unpad,
+            k_unpad,
+            v_unpad,
+            qv_unpad,
+            cu_seqlens_q,
+            cu_seqlens_k,
+            seqused_q,
+            seqused_k,
+            max_seqlen_q,
+            max_seqlen_k,
+            q,
+            k,
+            v,
+            qv,
+            output_pad_fn,
+            dq_pad_fn,
+            dk_pad_fn,
+        ) = generate_qkv(q, k, v, query_padding_mask, key_padding_mask, qv=qv, kvpacked=False,
+                        query_unused_mask=query_unused_mask, key_unused_mask=key_unused_mask)
+        q_unpad, k_unpad, v_unpad = [x.detach().to(dtype).requires_grad_() for x in (q_unpad, k_unpad, v_unpad)]
+        out_ref, attn_ref = attention_ref(
+            q_ref,
+            k_ref,
+            v_ref,
+            query_padding_mask,
+            key_padding_mask,
+            causal=causal,
+            qv=qv_ref,
+            q_descale=q_descale, k_descale=k_descale, v_descale=v_descale,
+            window_size=window_size,
+            attention_chunk=attention_chunk,
+            softcap=softcap
+        )
+        out_pt, attn_pt = attention_ref(
+            q_ref,
+            k_ref,
+            v_ref,
+            query_padding_mask,
+            key_padding_mask,
+            causal=causal,
+            qv=qv_ref,
+            q_descale=q_descale, k_descale=k_descale, v_descale=v_descale,
+            window_size=window_size,
+            attention_chunk=attention_chunk,
+            softcap=softcap,
+            upcast=False,
+            reorder_ops=True,
+            intermediate_dtype=dtype if dtype == torch.float8_e4m3fn else None,
+        )
+
+
+        print(f"Pytorch max diff: {(out_pt - out_ref).abs().max().item()}")
+        print(f"Pytorch mean diff: {(out_pt - out_ref).abs().mean().item()}")
+
+        if query_unused_mask is not None:
+            q_zero_masking = rearrange(query_unused_mask, "b s -> b s 1 1")
+
+        # Numerical error if we just do any arithmetic on out_ref
+        fwd_atol = 2 * (out_ref + 0.3 - 0.3 - out_ref).abs().max().item()
+        rtol = 2 if softcap == 0.0 else 3
+
+        pack_gqa_vals = [False, True] if not DISABLE_PACKGQA else [False]
+        num_splits_vals = [1, 3] if not DISABLE_SPLIT else [1]
+        for pack_gqa, num_splits in itertools.product(pack_gqa_vals, num_splits_vals):
+            out_unpad = flash_attn_varlen_func(
+                q_unpad,
+                k_unpad,
+                v_unpad,
+                cu_seqlens_q,
+                cu_seqlens_k,
+                max_seqlen_q,
+                max_seqlen_k,
+                seqused_q=seqused_q,
+                seqused_k=seqused_k,
+                causal=causal,
+                qv=qv_unpad,
+                q_descale=q_descale,
+                k_descale=k_descale, v_descale=v_descale,
+                window_size=window_size,
+                attention_chunk=attention_chunk,
+                softcap=softcap,
+            )
+            out = output_pad_fn(out_unpad)
+            if query_unused_mask is not None:
+                out.masked_fill_(q_zero_masking, 0.0)
+            print(f"Output max diff: {(out - out_ref).abs().max().item()}")
+            print(f"Output mean diff: {(out - out_ref).abs().mean().item()}")
+            # if not causal:
+            #     print(f"LSE max diff: {(lse - lse_ref).abs().max().item()}")
+            # breakpoint()
+
+            # Check that FlashAttention's numerical error is at most 3x the numerical error
+            # of a Pytorch implementation.
+            assert (out - out_ref).abs().max().item() <= rtol * (out_pt - out_ref).abs().max().item() + fwd_atol
+
+
+        if (
+            not DISABLE_BACKWARD 
+            and dtype != torch.float8_e4m3fn 
+            and not has_qv
+            and not dv > 256
+            and not attention_chunk != 0
+        ):
+            g_unpad = torch.randn_like(out_unpad)
+            do_o = ((g_unpad.float() * out_unpad.float()).sum(-1)).transpose(-1, -2)
+            # import flash_attn_3_cuda
+            # dq_unpad, dk_unpad, dv_unpad, softmax_d, dq_accum, lse_log2 = flash_attn_3_cuda.bwd_varlen(
+            #     g_unpad,
+            #     q_unpad,
+            #     k_unpad,
+            #     v_unpad,
+            #     out_unpad,
+            #     lse,
+            #     None,
+            #     None,
+            #     None,
+            #     cu_seqlens_q,
+            #     cu_seqlens_k,
+            #     None, None,
+            #     max_seqlen_q,
+            #     max_seqlen_k,
+            #     d ** (-0.5),
+            #     causal,
+            #     window_size[0], window_size[1],
+            #     softcap,
+            #     deterministic,
+            #     0,  # sm_margin
+            # )
+            dq_unpad, dk_unpad, dv_unpad = torch.autograd.grad(out_unpad, (q_unpad, k_unpad, v_unpad), g_unpad)
+            dq = dq_pad_fn(dq_unpad)
+            dk = dk_pad_fn(dk_unpad)
+            dv = dk_pad_fn(dv_unpad)
+            if key_unused_mask is not None:
+                k_zero_masking = rearrange(key_unused_mask, "b s -> b s 1 1")
+                dk.masked_fill_(k_zero_masking, 0.0)
+                dv.masked_fill_(k_zero_masking, 0.0)
+            if query_unused_mask is not None:
+                dq.masked_fill_(q_zero_masking, 0.0)
+            # print(f"dO_O max diff: {(softmax_d - do_o).abs().max().item()}")
+            # assert (softmax_d - do_o).abs().max().item() <= 1e-5
+            # assert dq_accum.abs().max().item() == 0.0
+            g = output_pad_fn(g_unpad)
+
+            # qk = torch.einsum('bthd,bshd->bhts', q / (d ** 0.5), k).float()
+            # qk = torch.masked_fill(qk, rearrange(~key_padding_mask, "b s -> b 1 1 s"), float("-inf"))
+            # dS = torch.einsum('bthd,bshd->bhts', g.float(), v.float())
+            # P = torch.softmax(qk, -1)
+            # dP = P * (dS - (g.float() * out.float()).sum(-1).transpose(1, 2).unsqueeze(-1))
+            # dQ = torch.einsum('bhts,bshd->bthd', dP, k.float())
+            # dV = torch.einsum('bhts,bthd->bshd', P, g.float())
+            # dK = torch.einsum('bhts,bthd->bshd', dP, q.float())
+
+
+            # dq, dk, dv = torch.autograd.grad(out, (q, k, v), g)
+            dq_ref, dk_ref, dv_ref = torch.autograd.grad(out_ref, (q_ref, k_ref, v_ref), g)
+            dq_pt, dk_pt, dv_pt = torch.autograd.grad(out_pt, (q_ref, k_ref, v_ref), g)
+            print(f"dQ max diff: {(dq - dq_ref).abs().max().item()}")
+            print(f"dK max diff: {(dk - dk_ref).abs().max().item()}")
+            print(f"dV max diff: {(dv - dv_ref).abs().max().item()}")
+            print(f"dQ mean diff: {(dq - dq_ref).abs().mean().item()}")
+            print(f"dK mean diff: {(dk - dk_ref).abs().mean().item()}")
+            print(f"dV mean diff: {(dv - dv_ref).abs().mean().item()}")
+            print(f"dQ Pytorch max diff: {(dq_pt - dq_ref).abs().max().item()}")
+            print(f"dK Pytorch max diff: {(dk_pt - dk_ref).abs().max().item()}")
+            print(f"dV Pytorch max diff: {(dv_pt - dv_ref).abs().max().item()}")
+            print(f"dQ Pytorch mean diff: {(dq_pt - dq_ref).abs().mean().item()}")
+            print(f"dK Pytorch mean diff: {(dk_pt - dk_ref).abs().mean().item()}")
+            print(f"dV Pytorch mean diff: {(dv_pt - dv_ref).abs().mean().item()}")
+            # breakpoint()
+            dq_atol = 2 * (dq_ref + 0.3 - 0.3 - dq_ref).abs().max().item() + (0 if softcap == 0 else 3e-4)
+            assert (dq - dq_ref).abs().max().item() <= rtol * (dq_pt - dq_ref).abs().max().item() + dq_atol
+            dk_atol = 2 * (dk_ref + 0.3 - 0.3 - dk_ref).abs().max().item() + (0 if softcap == 0 else 3e-4)
+            assert (dk - dk_ref).abs().max().item() <= rtol * (dk_pt - dk_ref).abs().max().item() + dk_atol
+            dv_atol = 2 * (dv_ref + 0.3 - 0.3 - dv_ref).abs().max().item() + (0 if softcap == 0 else 3e-4)
+            assert (dv - dv_ref).abs().max().item() <= rtol * (dv_pt - dv_ref).abs().max().item() + dv_atol
+
+
+# @pytest.mark.parametrize("dtype", [torch.float16, torch.bfloat16, torch.float8_e4m3fn])
+@pytest.mark.parametrize("dtype", [torch.bfloat16] + ([torch.float8_e4m3fn] if not DISABLE_FP8 else []))
+# @pytest.mark.parametrize("dtype", [torch.bfloat16])
+# @pytest.mark.parametrize("dtype", [torch.float8_e4m3fn])
+@pytest.mark.parametrize("mha_type", ["mha", "mqa", "gqa"])
+# @pytest.mark.parametrize("mha_type", ["mha"])
+@pytest.mark.parametrize("new_kv", [False] + ([True] if not DISABLE_APPENDKV else []))
+# @pytest.mark.parametrize("new_kv", [True])
+@pytest.mark.parametrize("causal,local", [(False, False), (True, False)] + ([(False, True)] if not DISABLE_LOCAL else []))
+# @pytest.mark.parametrize("causal,local", [(False, False), (True, False)])
+# @pytest.mark.parametrize("causal,local", [(False, False)])
+@pytest.mark.parametrize("seqlen_new_eq_seqlen_q", [True, False] if not DISABLE_APPENDKV else [True])
+# @pytest.mark.parametrize("seqlen_new_eq_seqlen_q", [True])
+@pytest.mark.parametrize("has_rotary_seqlens", [False, True])
+# @pytest.mark.parametrize("has_rotary_seqlens", [False])
+@pytest.mark.parametrize("rotary_interleaved", [False, True] if not DISABLE_APPENDKV else [False])
+# @pytest.mark.parametrize("rotary_interleaved", [True])
+@pytest.mark.parametrize("rotary_fraction", [0.0, 0.5, 1.0] if (not DISABLE_APPENDKV) and (apply_rotary_emb is not None) else [0.0])
+# @pytest.mark.parametrize("rotary_fraction", [0.0])
+@pytest.mark.parametrize("page_size", [None] + ([1, 4, 128] if not DISABLE_PAGEDKV else []))
+# @pytest.mark.parametrize("page_size", [None])
+@pytest.mark.parametrize("has_leftpad", [False, True])
+# @pytest.mark.parametrize("has_leftpad", [False])
+@pytest.mark.parametrize("has_batch_idx", [False, True])
+# @pytest.mark.parametrize("has_batch_idx", [False])
+@pytest.mark.parametrize("varlen_q", [False, True])
+# @pytest.mark.parametrize("varlen_q", [False])
+# @pytest.mark.parametrize("d", [32, 59, 64, 80, 128, 256])
+# @pytest.mark.parametrize("d", [32, 64, 96, 128, 160, 192, 224, 256])
+# @pytest.mark.parametrize('d', [32, 40, 64, 80, 96, 128, 160, 192])
+# @pytest.mark.parametrize('d', [56, 80])
+@pytest.mark.parametrize("d", [128])
+# @pytest.mark.parametrize("d", [192])
+@pytest.mark.parametrize(
+    "seqlen_q,seqlen_k",
+    [
+        (1, 128),
+        (1, 339),
+        (3, 1024),
+        (64, 800),
+        (64, 256),
+        (3, 799),
+        (64, 2048),
+        (16, 20000),
+        # (1, 128 * 1024),
+        # (16, 128 * 1024),
+        (128, 128),
+        (256, 512),  # To test appending KV with more than 1 block
+        (2048, 3577),  # Enough tile to test persistent scheduler
+    ],
+)
+# @pytest.mark.parametrize('seqlen_q,seqlen_k', [(256, 128)])
+def test_flash_attn_kvcache(
+    seqlen_q,
+    seqlen_k,
+    d,
+    varlen_q,
+    has_batch_idx,
+    has_leftpad,
+    page_size,
+    rotary_fraction,
+    rotary_interleaved,
+    has_rotary_seqlens,
+    seqlen_new_eq_seqlen_q,
+    causal,
+    local,
+    new_kv,
+    mha_type,
+    dtype,
+):
+    if page_size is not None and seqlen_k % page_size != 0:
+        pytest.skip()
+    if seqlen_q > seqlen_k and new_kv:
+        pytest.skip()
+    if not new_kv and rotary_fraction > 0.0:
+        pytest.skip()
+    if rotary_fraction == 0.0 and has_rotary_seqlens:
+        pytest.skip()
+    device = "cuda"
+    # set seed
+    torch.random.manual_seed(0)
+    batch_size = 5
+    # batch_size = 1
+    batch_size_cache = batch_size if not has_batch_idx else batch_size * 2
+    nheads = 6
+    # nheads = 1
+    # rotary_dim must be a multiple of 16, and must be <= d
+    rotary_dim = math.floor(int(rotary_fraction * d) / 16) * 16
+    nheads_k = nheads if mha_type == "mha" else (1 if mha_type == "mqa" else 3)
+    assert nheads % nheads_k == 0
+    dtype_ref = torch.bfloat16 if dtype == torch.float8_e4m3fn else dtype
+    dv_vals = [128, d] if d > 128 and d <= 192 else ([256, 512, d] if d <= 64 else [d])
+    if dtype == torch.float8_e4m3fn:
+        dv_vals = [d]
+    attention_chunk_vals = [torch.randint(1, seqlen_k * 2, (1,)).item(), 0] if (causal or local) and not DISABLE_LOCAL else [0]
+    for dv, attention_chunk in itertools.product(dv_vals, attention_chunk_vals):
+        has_qv = d == 64 and dv >= 256
+        q = torch.randn(batch_size, seqlen_q, nheads, d, device=device, dtype=dtype_ref).to(dtype).to(dtype_ref)
+        if has_qv:
+            qv = torch.randn(batch_size, seqlen_q, nheads, dv, device=device, dtype=dtype_ref).to(dtype).to(dtype_ref)
+        else:
+            qv = None
+        if varlen_q:
+            query_padding_mask = generate_random_padding_mask(seqlen_q, batch_size, device, mode="random")
+            q_unpad, indices_q, cu_seqlens_q, max_seqlen_q, *rest = unpad_input(q, query_padding_mask)
+            output_pad_fn = lambda output_unpad: pad_input(
+                output_unpad, indices_q, batch_size, seqlen_q
+            )
+            qv_unpad = rearrange(qv, "b s ... -> (b s) ...")[indices_q] if has_qv else None
+        else:
+            query_padding_mask = None
+            q_unpad = q
+            qv_unpad = qv
+            cu_seqlens_q, max_seqlen_q = None, None
+        # Put window_size after QKV randn so that window_size changes from test to test
+        window_size = (-1, -1) if not local else torch.randint(0, seqlen_k, (2,))
+
+        seqlen_new = seqlen_q if seqlen_new_eq_seqlen_q else torch.randint(1, seqlen_q + 1, (1,)).item()
+        cu_seqlens_k_new = None
+        key_new_padding_mask = None
+        if new_kv:
+            k = torch.randn(batch_size, seqlen_new, nheads_k, d, device=device, dtype=dtype_ref).to(dtype).to(dtype_ref)
+            v = torch.randn(batch_size, seqlen_new, nheads_k, dv, device=device, dtype=dtype_ref).to(dtype).to(dtype_ref)
+            if varlen_q:  # k & v are also varlen
+                key_new_padding_mask = generate_random_padding_mask(seqlen_new, batch_size, device, mode="random")
+                k_unpad, indices_k, cu_seqlens_k_new, *rest = unpad_input(k, key_new_padding_mask)
+                v_unpad, *rest = unpad_input(v, key_new_padding_mask)
+            else:
+                k_unpad, v_unpad = k, v
+        else:
+            k, v, k_unpad, v_unpad = None, None, None, None
+        if page_size is None:
+            k_cache = torch.randn(batch_size_cache, seqlen_k, nheads_k, d, device=device, dtype=dtype_ref).to(dtype).to(dtype_ref)
+            v_cache = torch.randn(batch_size_cache, seqlen_k, nheads_k, dv, device=device, dtype=dtype_ref).to(dtype).to(dtype_ref)
+            page_table = None
+        else:
+            (
+                k_cache,
+                v_cache,
+                page_table,
+                k_cache_paged,
+                v_cache_paged,
+                num_blocks,
+            ) = _generate_block_kvcache(
+                seqlen_k, page_size, batch_size_cache, nheads_k, d, dv, device, dtype, dtype_ref
+            )
+        cache_seqlens = torch.randint(
+            0 if new_kv else 1,
+            # If we don't use seqlen_q in the case of causal and rotary, cos/sin won't be long enough
+            (
+                (seqlen_k - (seqlen_q if (causal or local) and rotary_dim > 1 else seqlen_new) + 1)
+                if new_kv
+                else (seqlen_k + 1)
+            ),
+            (batch_size,),
+            dtype=torch.int32,
+            device=device,
+        )
+        if has_leftpad:
+            cache_leftpad = torch.cat([torch.randint(0, cache_seqlens[i].item(), (1,), dtype=torch.int32, device=device)
+                                    if cache_seqlens[i].item() > 0 else torch.zeros(1, dtype=torch.int32, device=device)
+                                    for i in range(batch_size)])
+        else:
+            cache_leftpad = None
+        if has_batch_idx:
+            cache_batch_idx = torch.randperm(batch_size_cache, dtype=torch.int32, device=device)[
+                :batch_size
+            ]
+        else:
+            cache_batch_idx = None
+        arange = rearrange(torch.arange(seqlen_k, device=device), "s -> 1 s")
+        cache_seqlens_expanded = rearrange(cache_seqlens, "b -> b 1")
+        if not new_kv:
+            key_padding_mask = arange < cache_seqlens_expanded
+        else:
+            k_new_seqlens = key_new_padding_mask.sum(-1, keepdims=True) if varlen_q else seqlen_new
+            key_padding_mask = arange < cache_seqlens_expanded + k_new_seqlens
+        if has_leftpad:
+            key_padding_mask = torch.logical_and(
+                key_padding_mask, arange >= cache_leftpad.unsqueeze(-1).expand(-1, seqlen_k)
+            )
+        # cache_seqlens = torch.tensor([64], dtype=torch.int32, device=device)
+        rotary_seqlens = cache_seqlens if not has_rotary_seqlens else cache_seqlens // 2
+        if rotary_dim > 0:
+            angle = (
+                torch.rand(
+                    seqlen_k if page_size is None else num_blocks * page_size,
+                    rotary_dim // 2,
+                    device=device,
+                )
+                * 2
+                * math.pi
+            )
+            cos = torch.cos(angle).to(dtype=dtype_ref).to(dtype).to(dtype_ref)
+            sin = torch.sin(angle).to(dtype=dtype_ref).to(dtype).to(dtype_ref)
+            if causal or local:
+                q_ro = apply_rotary_emb(
+                    q, cos, sin, seqlen_offsets=rotary_seqlens, interleaved=rotary_interleaved
+                )
+            else:
+                q_ro = rearrange(
+                    apply_rotary_emb(
+                        rearrange(q, "b s h d -> b 1 (s h) d"),
+                        cos,
+                        sin,
+                        seqlen_offsets=rotary_seqlens,
+                        interleaved=rotary_interleaved,
+                    ),
+                    "b 1 (s h) d -> b s h d",
+                    s=seqlen_q,
+                )
+            # q_ro = q
+            k_ro = apply_rotary_emb(
+                k, cos, sin, seqlen_offsets=rotary_seqlens, interleaved=rotary_interleaved
+            )
+        else:
+            cos, sin = None, None
+            q_ro, k_ro = q, k
+        # k_cache[:, 64:] = -1
+        k_cache_ref = (k_cache if not has_batch_idx else k_cache[cache_batch_idx]).clone()
+        v_cache_ref = (v_cache if not has_batch_idx else v_cache[cache_batch_idx]).clone()
+        if new_kv:
+            update_mask = torch.logical_and(
+                cache_seqlens_expanded <= arange, arange < cache_seqlens_expanded + k_new_seqlens
+            )
+            k_to_update = rearrange(k_ro, "b s ... -> (b s) ...")
+            v_to_update = rearrange(v, "b s ... -> (b s) ...")
+            if varlen_q:
+                k_to_update = k_to_update[indices_k]
+                v_to_update = v_to_update[indices_k]
+            k_cache_ref[update_mask] = k_to_update
+            v_cache_ref[update_mask] = v_to_update
+        k_cache_rep = repeat(k_cache_ref, "b s h d -> b s (h g) d", g=nheads // nheads_k)
+        v_cache_rep = repeat(v_cache_ref, "b s h d -> b s (h g) d", g=nheads // nheads_k)
+        out_ref, _ = attention_ref(
+            q_ro,
+            k_cache_rep,
+            v_cache_rep,
+            query_padding_mask,
+            key_padding_mask,
+            causal=causal,
+            qv=qv,
+            window_size=window_size,
+            attention_chunk=attention_chunk,
+            key_leftpad=cache_leftpad,
+        )
+        out_pt, _ = attention_ref(
+            q_ro,
+            k_cache_rep,
+            v_cache_rep,
+            query_padding_mask,
+            key_padding_mask,
+            causal=causal,
+            qv=qv,
+            window_size=window_size,
+            attention_chunk=attention_chunk,
+            upcast=False,
+            reorder_ops=True,
+            key_leftpad=cache_leftpad,
+            intermediate_dtype=dtype if dtype == torch.float8_e4m3fn else None
+        )
+        q = q.to(dtype)
+        q_unpad = q_unpad.to(dtype) if varlen_q else None
+        k_cache = k_cache.to(dtype)
+        v_cache = v_cache.to(dtype)
+        k_cache_paged = k_cache_paged.to(dtype) if page_size is not None else None
+        v_cache_paged = v_cache_paged.to(dtype) if page_size is not None else None
+        k = k.to(dtype) if k is not None else None
+        v = v.to(dtype) if v is not None else None
+        k_unpad = k_unpad.to(dtype) if k_unpad is not None else None
+        v_unpad = v_unpad.to(dtype) if v_unpad is not None else None
+        qv = qv.to(dtype) if qv is not None else None
+        qv_unpad = qv_unpad.to(dtype) if (varlen_q and qv is not None) else None
+        cos = cos.to(dtype) if cos is not None else None
+        sin = sin.to(dtype) if sin is not None else None
+        k_cache_saved = k_cache.clone() if page_size is None else k_cache_paged.clone()
+        v_cache_saved = v_cache.clone() if page_size is None else v_cache_paged.clone()
+        num_splits_vals = [1, 0] if not DISABLE_SPLIT else [1]
+        precompute_metadata_vals = [False, True]
+        for num_splits, precompute_metadata in itertools.product(num_splits_vals, precompute_metadata_vals):
+            if precompute_metadata:
+                scheduler_metadata = get_scheduler_metadata(
+                    batch_size, max_seqlen_q if varlen_q else seqlen_q, seqlen_k, nheads, nheads_k, d,
+                    cache_seqlens, q.dtype, headdim_v=dv, cu_seqlens_q=cu_seqlens_q,
+                    cu_seqlens_k_new=cu_seqlens_k_new, cache_leftpad=cache_leftpad,
+                    max_seqlen_k_new=seqlen_new, page_size=page_size,
+                    causal=causal, window_size=window_size, attention_chunk=attention_chunk,
+                    num_splits=num_splits
+                )
+            else:
+                scheduler_metadata = None
+            # Repeat to test metadata reuse
+            for _ in range(1 if not precompute_metadata else 2):
+                if page_size is None:
+                    k_cache.copy_(k_cache_saved)
+                    v_cache.copy_(v_cache_saved)
+                else:
+                    k_cache_paged.copy_(k_cache_saved)
+                    v_cache_paged.copy_(v_cache_saved)
+                out, lse, *rest = flash_attn_with_kvcache(
+                    q if not varlen_q else q_unpad,
+                    k_cache if page_size is None else k_cache_paged,
+                    v_cache if page_size is None else v_cache_paged,
+                    k if not new_kv or not varlen_q else k_unpad,
+                    v if not new_kv or not varlen_q else v_unpad,
+                    qv=qv if not varlen_q else qv_unpad,
+                    rotary_cos=cos,
+                    rotary_sin=sin,
+                    cache_seqlens=cache_seqlens,
+                    cache_batch_idx=cache_batch_idx,
+                    cache_leftpad=cache_leftpad,
+                    page_table=page_table,
+                    cu_seqlens_q=cu_seqlens_q,
+                    cu_seqlens_k_new=cu_seqlens_k_new,
+                    max_seqlen_q=max_seqlen_q,
+                    rotary_seqlens=rotary_seqlens,
+                    causal=causal,
+                    window_size=window_size,
+                    attention_chunk=attention_chunk,
+                    rotary_interleaved=rotary_interleaved,
+                    scheduler_metadata=scheduler_metadata,
+                    num_splits=num_splits,
+                    return_softmax_lse=True
+                )
+                if varlen_q:
+                    out = output_pad_fn(out)
+                # out = flash_attn_with_kvcache(
+                #     q, k_cache, v_cache, cache_seqlens=cache_seqlens, causal=causal, window_size=window_size
+                # )
+                # out = flash_attn_with_kvcache(q, k_cache, v_cache, causal=causal, window_size=window_size)
+                # qk = torch.einsum("bqhd,bkhd->bhqk", q, k_cache_ref)
+                # m = qk.amax(-1, keepdim=True)
+                # s_tmp = torch.exp((qk - m) / math.sqrt(d))
+                # o1 = torch.einsum('bhst,bthd->bshd', s_tmp, v_cache_ref)
+                # lse_ref = torch.logsumexp(qk / math.sqrt(d), -1)
+                # probs = torch.softmax(qk, dim=-1)
+                print(f"Output max diff: {(out - out_ref).abs().max().item()}")
+                print(f"Output mean diff: {(out - out_ref).abs().mean().item()}")
+                print(f"Pytorch max diff: {(out_pt - out_ref).abs().max().item()}")
+                print(f"Pytorch mean diff: {(out_pt - out_ref).abs().mean().item()}")
+                # breakpoint()
+
+                # Check that FlashAttention's numerical error is at most twice the numerical error
+                # of a Pytorch implementation.
+                if new_kv:
+                    if page_size is None:
+                        k_cache_select = (
+                            k_cache.to(dtype_ref) if not has_batch_idx else k_cache.to(dtype_ref)[cache_batch_idx]
+                        )
+                        v_cache_select = (
+                            v_cache.to(dtype_ref) if not has_batch_idx else v_cache.to(dtype_ref)[cache_batch_idx]
+                        )
+                    else:
+                        k_cache_select = rearrange(
+                            k_cache_paged.to(dtype_ref)[(page_table if not has_batch_idx else page_table[cache_batch_idx]).flatten()],
+                            "(b nblocks) block_size ... -> b (nblocks block_size) ...",
+                            b=batch_size,
+                        )[:, :seqlen_k].to(dtype_ref)
+                        v_cache_select = rearrange(
+                            v_cache_paged.to(dtype_ref)[(page_table if not has_batch_idx else page_table[cache_batch_idx]).flatten()],
+                            "(b nblocks) block_size ... -> b (nblocks block_size) ...",
+                            b=batch_size,
+                        )[:, :seqlen_k].to(dtype_ref)
+                    k_cache_ref = k_cache_ref.to(dtype).to(dtype_ref)
+                    v_cache_ref = v_cache_ref.to(dtype).to(dtype_ref)
+                    if dtype is not torch.float8_e4m3fn:
+                        assert torch.equal(v_cache_select, v_cache_ref)
+                    else:
+                        assert torch.allclose(v_cache_select, v_cache_ref, rtol=1e-3, atol=1e-3)
+                    # breakpoint()
+                    # if rotary_dim == 0 and dtype is not torch.float8_e4m3fn:
+                    if rotary_dim == 0:
+                        assert torch.equal(k_cache_select, k_cache_ref)
+                    else:
+                        # if not torch.allclose(k_cache_select, k_cache_ref, rtol=1e-3, atol=1e-3):
+                        #     breakpoint()
+                        if dtype is not torch.float8_e4m3fn:
+                            assert torch.allclose(k_cache_select, k_cache_ref, rtol=1e-3, atol=1e-3)
+                        else:
+                            assert torch.allclose(k_cache_select, k_cache_ref, rtol=1e-1, atol=1e-1)
+                mult = 4 if dtype == torch.float8_e4m3fn else 2
+                assert (out - out_ref).abs().max().item() <= mult * (out_pt - out_ref).abs().max().item() + 1e-5
+                mult_mean = 3 if dtype == torch.float8_e4m3fn else 1.5
+                assert (out - out_ref).abs().mean().item() <= mult_mean * (out_pt - out_ref).abs().mean().item()
+
+
+def _generate_block_kvcache(seqlen_k, page_size, batch_size, nheads_k, d, dv, device, dtype, dtype_ref):
+    num_blocks = math.ceil(seqlen_k / page_size) * batch_size * 3
+    k_cache_paged = torch.randn(
+        num_blocks, page_size, nheads_k, d, device=device, dtype=dtype_ref
+    ).to(dtype).to(dtype_ref)
+    v_cache_paged = torch.randn(
+        num_blocks, page_size, nheads_k, dv, device=device, dtype=dtype_ref
+    ).to(dtype).to(dtype_ref)
+    page_table = rearrange(
+        torch.randperm(num_blocks, dtype=torch.int32, device=device),
+        "(b nblocks) -> b nblocks",
+        b=batch_size,
+    )
+    k_cache = rearrange(
+        k_cache_paged[page_table.flatten()],
+        "(b nblocks) block_size ... -> b (nblocks block_size) ...",
+        b=batch_size,
+    )[:, :seqlen_k]
+    v_cache = rearrange(
+        v_cache_paged[page_table.flatten()],
+        "(b nblocks) block_size ... -> b (nblocks block_size) ...",
+        b=batch_size,
+    )[:, :seqlen_k]
+    return k_cache, v_cache, page_table, k_cache_paged, v_cache_paged, num_blocks
+
+
+@pytest.mark.parametrize("dtype", [torch.bfloat16])
+@pytest.mark.parametrize("causal", [False, True])
+# @pytest.mark.parametrize('causal', [False])
+@pytest.mark.parametrize('d', [128])
+@pytest.mark.parametrize(
+    "seqlen_q,seqlen_k",
+    [
+        (64, 8192),
+    ],
+)
+def test_flash_attn_cluster(seqlen_q, seqlen_k, d, causal, dtype):
+    device = "cuda"
+    torch.random.manual_seed(0)
+    batch_size = 2
+    nheads = 16
+    nheads_kv = 4
+    # There was a bug where this would cause "unspecified launch failure" due to Cluster
+    q = torch.randn(batch_size, seqlen_q, nheads, d, device=device, dtype=dtype)
+    k = torch.randn(batch_size, seqlen_k, nheads_kv, d, device=device, dtype=dtype)
+    v = torch.randn(batch_size, seqlen_k, nheads_kv, d, device=device, dtype=dtype)
+    for _ in range(100):
+        flash_attn_func(q, k, v, causal=causal)
+
+
+# @pytest.mark.parametrize("dtype", ([torch.float16] if is_sm75 else [torch.float16, torch.bfloat16]))
+@pytest.mark.parametrize("dtype", [torch.bfloat16])
+@pytest.mark.parametrize("causal", [False, True])
+# @pytest.mark.parametrize('causal', [False])
+@pytest.mark.parametrize("d", [32, 40, 59, 64, 80, 96, 111, 128, 160, 192, 224, 256])
+# @pytest.mark.parametrize("d", [32, 40, 59, 64, 80, 96, 111, 128])
+# @pytest.mark.parametrize('d', [32, 56, 64, 80, 96, 128])
+# @pytest.mark.parametrize("d", [32, 64, 96, 128, 160, 192])
+# @pytest.mark.parametrize('d', [80])
+@pytest.mark.parametrize(
+    "seqlen_q,seqlen_k",
+    [
+        (1, 239),
+        (239, 1),
+        (3, 799),
+        (799, 3),
+        (1024, 128),
+        (97, 97),
+        (128, 128),
+        (200, 200),
+        (256, 256),
+        (257, 257),
+        (384, 384),
+        (512, 512),
+        (768, 768),
+        (1024, 1024),
+        (2048, 2048),
+    ],
+)
+def test_flash_attn_race_condition(seqlen_q, seqlen_k, d, causal, dtype):
+    device = "cuda"
+    # set seed
+    torch.random.manual_seed(0)
+    # Simulate under memory load
+    dummy = torch.empty(70 * 1024 ** 3, dtype=torch.uint8, device=device)
+    batch_size = 60  # Sometimes we need large batch size for the race conditions to trigger
+    nheads = 4
+    q = torch.randn(batch_size, seqlen_q, nheads, d, device=device, dtype=dtype, requires_grad=True)
+    k = torch.randn(batch_size, seqlen_k, nheads, d, device=device, dtype=dtype, requires_grad=True)
+    v = torch.randn(batch_size, seqlen_k, nheads, d, device=device, dtype=dtype, requires_grad=True)
+    torch.random.manual_seed(42)
+    out0 = flash_attn_func(q, k, v, causal=causal)
+    g = torch.randn_like(out0)
+    dq0, dk0, dv0 = torch.autograd.grad(out0, (q, k, v), g)
+    # Numerical error if we just do any arithmetic on dq
+    dq_atol = 2 * ((dq0 + 0.3 - 0.3) - dq0).abs().max().item()
+
+    for i in range(1000):
+        torch.random.manual_seed(42)
+        out = flash_attn_func(q, k, v, causal=causal)
+        assert torch.equal(out, out0)
+        # assert torch.equal(lse, lse0)
+
+        dq, dk, dv = torch.autograd.grad(out, (q, k, v), g)
+        dq_equal = torch.allclose(dq, dq0, atol=dq_atol)
+        if not dq_equal:
+            print(f"Iter {i}, {dq_atol = }, dQ max diff: {(dq - dq0).abs().max().item()}")
+            # breakpoint()
+        assert torch.equal(dv, dv0)
+        assert torch.equal(dk, dk0)
+        assert dq_equal
+
+
+def attention_combine_ref(out_partial, lse_partial):
+    """
+    out_partial: (num_splits, batch_size, seqlen, nheads, d)
+    lse_partial: (num_splits, batch_size, nheads, seqlen)
+    """
+    lse = torch.logsumexp(lse_partial, dim=0)
+    scale = torch.exp(lse_partial - lse)
+    scale = torch.where(torch.isinf(scale) | torch.isnan(scale), torch.zeros_like(scale), scale)
+    out = (scale.unsqueeze(-1) * out_partial).sum(0)
+    return out, lse
+
+
+@pytest.mark.parametrize("dtype", [torch.float32, torch.float16, torch.bfloat16])
+# @pytest.mark.parametrize("dtype", [torch.float32])
+# @pytest.mark.parametrize("d", [32, 40, 59, 64, 80, 96, 111, 128, 160, 192, 224, 256])
+@pytest.mark.parametrize("d", [64, 96, 128, 192, 256, 512])
+# @pytest.mark.parametrize("d", [128])
+@pytest.mark.parametrize("seqlen", [1, 2, 3, 32, 64, 256, 113, 108, 640, 1024])
+# @pytest.mark.parametrize("seqlen", [12, 32, 64, 256, 112, 108, 640, 1024, 2048, 8192])
+# @pytest.mark.parametrize("seqlen", [15])
+@pytest.mark.parametrize("num_splits", [1, 2, 3, 5, 17, 32, 55, 97, 133])
+# @pytest.mark.parametrize("num_splits", [1, 2, 3, 5, 11])
+# @pytest.mark.parametrize("num_splits", [128])
+def test_flash_attn_combine(num_splits, seqlen, d, dtype):
+    if DISABLE_SPLIT:
+        pytest.skip()
+    device = "cuda"
+    # set seed
+    torch.random.manual_seed(1)
+    batch_size = 5
+    nheads = 16
+    # batch_size = 1
+    # nheads = 1
+    out_partial = torch.randn(num_splits * 2, batch_size, nheads, seqlen, d, device=device, dtype=torch.float32).transpose(2, 3)[:num_splits]  # To test non-contiguous tensor
+    lse_partial = torch.randn(num_splits, batch_size, nheads * 2, seqlen, device=device, dtype=torch.float32).transpose(-1, -2)[:, :, :, :nheads]  # To test non-contiguous tensor
+    # To test short-circuiting based on num_splits
+    lse_partial[num_splits // 2:, :batch_size // 3] = -float("inf")
+    out, lse = flash_attn_combine(out_partial, lse_partial, out_dtype=dtype)
+    out_ref, lse_ref = attention_combine_ref(out_partial, lse_partial)
+    out_pt = out_ref.to(dtype)
+
+    print(f"LSE max diff: {(lse - lse_ref).abs().max().item()}")
+    print(f"LSE mean diff: {(lse - lse_ref).abs().mean().item()}")
+    print(f"Output max diff: {(out - out_ref).abs().max().item()}")
+    print(f"Output mean diff: {(out - out_ref).abs().mean().item()}")
+    print(f"Pytorch max diff: {(out_pt - out_ref).abs().max().item()}")
+    print(f"Pytorch mean diff: {(out_pt - out_ref).abs().mean().item()}")
+    # breakpoint()
+
+    assert torch.allclose(lse, lse_ref, atol=1e-5, rtol=1e-5)
+    multiple = 2
+    assert ((out - out_ref).abs().max().item() <= multiple * (out_pt - out_ref).abs().max().item()) or torch.allclose(out, out_pt, atol=1e-5, rtol=1e-5)
+
+    # from flash_attn.utils.benchmark import pytorch_profiler
+    # # pytorch_profiler(torch.sum, lse_partial)
+    # pytorch_profiler(flash_attn_combine, out_partial, lse_partial)
+    # pytorch_profiler(torch.sum, out_partial)
+
+def test_flash3_bw_compatibility() -> None:
+    # Let's try to always stay backward compatible! This will make life easier
+    # for downstream libaries, users, and exported models.
+    # 1/ Instead of removing arguments, error out if their value is no longer supported
+    # 2/ When adding arguments, add them at the end with a default value
+    assert torch.ops.flash_attn_3.fwd.default._schema.is_backward_compatible_with(parse_schema(
+        "flash_attn_3::fwd(Tensor q, Tensor k, Tensor v, Tensor(k_new!)? k_new=None, "
+        "Tensor(v_new!)? v_new=None, Tensor? q_v=None, Tensor(out!)? out=None, "
+        "Tensor? cu_seqlens_q=None, Tensor? cu_seqlens_k=None, "
+        "Tensor? cu_seqlens_k_new=None, Tensor? seqused_q=None, Tensor? seqused_k=None, "
+        "int? max_seqlen_q=None, int? max_seqlen_k=None, Tensor? page_table=None, "
+        "Tensor? kv_batch_idx=None, Tensor? leftpad_k=None, Tensor? rotary_cos=None, Tensor? rotary_sin=None, "
+        "Tensor? seqlens_rotary=None, Tensor? q_descale=None, Tensor? k_descale=None, Tensor? v_descale=None, "
+        "float? softmax_scale=None, bool is_causal=False, int window_size_left=-1, int window_size_right=-1, "
+        "int attention_chunk=0, float softcap=0., bool is_rotary_interleaved=False, "
+        "Tensor? scheduler_metadata=None, int num_splits=0, bool? pack_gqa=None, int sm_margin=0) "
+        "-> (Tensor(out!), Tensor, Tensor, Tensor)"
+    ))
+    assert torch.ops.flash_attn_3.bwd.default._schema.is_backward_compatible_with(parse_schema(
+        "flash_attn_3::bwd(Tensor dout, Tensor q, Tensor k, Tensor v, Tensor out, Tensor softmax_lse, "
+        "Tensor(dq!)? dq=None, Tensor(dk!)? dk=None, Tensor(dv!)? dv=None, Tensor? cu_seqlens_q=None, "
+        "Tensor? cu_seqlens_k=None, Tensor? seqused_q=None, Tensor? seqused_k=None, int? max_seqlen_q=None, "
+        "int? max_seqlen_k=None, float? softmax_scale=None, bool is_causal=False, int window_size_left=-1, "
+        "int window_size_right=-1, float softcap=0., bool deterministic=False, int sm_margin=0) "
+        "-> (Tensor(dq!), Tensor(dk!), Tensor(dv!), Tensor, Tensor, Tensor, Tensor, Tensor)"
+    ))
+    assert torch.ops.flash_attn_3.fwd_combine.default._schema.is_backward_compatible_with(parse_schema(
+        "flash_attn_3::fwd_combine(Tensor out_partial, Tensor lse_partial, Tensor(out!)? out=None, "
+        "ScalarType? out_dtype=None) -> (Tensor(out!), Tensor)"
+    ))
+    assert torch.ops.flash_attn_3.get_scheduler_metadata.default._schema.is_backward_compatible_with(parse_schema(
+        "flash_attn_3::get_scheduler_metadata(int batch_size, int max_seqlen_q, int max_seqlen_k, "
+        "int num_heads, int num_heads_k, int headdim, int headdim_v, ScalarType qkv_dtype, Tensor seqused_k, "
+        "Tensor? cu_seqlens_q=None, Tensor? cu_seqlens_k=None, Tensor? cu_seqlens_k_new=None, "
+        "Tensor? seqused_q=None, Tensor? leftpad_k=None, int? page_size=None, int max_seqlen_k_new=0, "
+        "bool is_causal=False, int window_size_left=-1, int window_size_right=-1, "
+        "int attention_chunk=0, bool has_softcap=False, int num_splits=0, bool? pack_gqa=None, "
+        "int sm_margin=0) -> Tensor"
+    ))

Code/Baselines/flash-attention/hopper/tile_size.h

@@ -0,0 +1,78 @@
+/******************************************************************************
+ * Copyright (c) 2024, Jay Shah, Ganesh Bikshandi, Ying Zhang, Vijay Thakkar, Pradeep Ramani, Tri Dao.
+ ******************************************************************************/
+
+#pragma once
+
+#include <tuple>
+
+// Return {kBlockM, kBlockN, MmaPV_is_RS, IntraWGOverlap}
+constexpr std::tuple<int, int, bool, bool> tile_size_fwd_sm90(
+        int headdim, int headdim_v, bool is_causal, bool is_local, int element_size=2,
+        bool v_colmajor=false, bool paged_kv_non_TMA=false, bool softcap=false) {
+    if (element_size == 2) {
+        if (headdim <= 64) {
+            // return {same_hdim ? 192 : 64, same_hdim ? 128 : 64, same_hdim, same_hdim};
+            // With this workaround in Cutlass 3.8, tile size 192 x 128 got slower for non-causal, idk why
+            // https://github.com/NVIDIA/cutlass/blob/833f6990e031b48b4cd2fcf55e0849c51ef6bac2/include/cute/container/tuple.hpp#L131
+            if (headdim_v == 512) {
+                return {64, 64, false, false};
+            } else if (headdim_v == 256) {
+                return {128, 96, true, false};
+            } else {
+                // Switch to tile size 192 x 192 for now
+                bool const use_blockN_128 = is_causal || is_local;
+                return {192, use_blockN_128 ? 128 : 192, use_blockN_128, true};
+            }
+            // Good for long seqlen (>= 4k) but suffers from tile quantization at short seqlen
+            // return {192, is_causal || is_local ? 192 : 176, true, false};
+        } else if (headdim <= 96) {
+            return {192, is_local || paged_kv_non_TMA ? 128 : 144, false, true};
+        } else if (headdim <= 128) {
+            return {128, is_causal || is_local || paged_kv_non_TMA ? 128 : 176, true, true};
+            // {128, 192, false, false} and {192, 128, false, true} are quite good too
+            // 128 x 192 hits the limit of smem if MmaPV_is_RS, 128 x 144 hits the limit if !MmaPV_is_RS
+        } else if (headdim <= 192) {
+            return {128, paged_kv_non_TMA || is_local ? 96 : (headdim_v <= 128 ? 128 : 112), true, true};  // 128 x 112 hits the limit of smem
+        } else {
+            return {128, is_local ? 64 : 80, true, true};  // 128 x 80 hits the limit of smem
+        }
+    } else {
+        if (headdim <= 64) {
+            return {192, 160, true, true};
+        } else if (headdim <= 96) {
+            return {192, 128, true, true};
+        } else if (headdim <= 128) {
+            return {128, paged_kv_non_TMA ? 160 : (v_colmajor || (softcap && is_local) ? 192 : 224), true, true};
+        } else if (headdim <= 192) {
+            return {128, (paged_kv_non_TMA || softcap) && is_local ? 128 : 160, true, true};
+        } else {
+            return {128, is_local ? 64 : 128, true, !paged_kv_non_TMA};  // PagedKV uses more registers so we disabled IntraWGOverlap
+        }
+    }
+}
+
+// Return {kBlockM, kBlockN, kNWarps, kStages, Q_in_regs}
+constexpr std::tuple<int, int, int, int, bool> tile_size_fwd_sm8x(
+        bool sm86_or_89, int headdim, int headdim_v, bool is_causal, bool is_local, int element_size=2,
+        bool paged_kv=false, bool varlen_and_split=false,
+        bool softcap=false, bool append_kv=false) {
+    if (element_size == 2) {
+        if (headdim <= 64) {
+            return {128, varlen_and_split ? 80 : (is_local ? 96 : 112), 4, 1, false};
+        } else if (headdim <= 96) {
+            return {128, varlen_and_split || is_local ? 48 : 64, 4, 1, false};
+        } else if (headdim <= 128) {
+            bool const use_8_warps = sm86_or_89 | varlen_and_split;
+            return {128, use_8_warps ? (varlen_and_split ? (is_local ? 96 : 112) : (is_local ? 96 : 128)) : (is_local ? 48 : 64), use_8_warps ? 8 : 4, 1, use_8_warps};
+        } else if (headdim <= 192) {
+            bool const kBlockN_64 = append_kv || is_local || varlen_and_split || paged_kv;
+            return {128, kBlockN_64 ? 64 : 96, 8, sm86_or_89 ? 1 : 2, !kBlockN_64};
+        } else {
+            return {128, sm86_or_89 ? (append_kv ? 32 : (varlen_and_split || is_local ? 48 : 64)) : (append_kv ? 48 : (varlen_and_split || is_local ? 64 : 96)), 8, 1, sm86_or_89 && !append_kv};
+        }
+    } else {
+        // Placeholder for now
+        return {128, 64, 8, 2, false};
+    }
+}

Code/Baselines/flash-attention/tests/test_flash_attn_ck.py

@@ -0,0 +1,1618 @@
+import math
+
+import pytest
+import torch
+import torch.nn.functional as F
+from einops import rearrange, repeat
+from flash_attn import (
+    flash_attn_func,
+    flash_attn_kvpacked_func,
+    flash_attn_qkvpacked_func,
+    flash_attn_varlen_func,
+    flash_attn_varlen_kvpacked_func,
+    flash_attn_varlen_qkvpacked_func,
+    flash_attn_with_kvcache,
+)
+
+from test_flash_attn import (
+    attn_bias_from_alibi_slopes,
+    convert_flash_attn_S_to_softmax,
+    generate_qkv,
+    generate_random_padding_mask,
+    _generate_block_kvcache,
+    attention_ref,
+    attention_kvpacked_ref,
+    attention_qkvpacked_ref,
+)
+
+from flash_attn.layers.rotary import apply_rotary_emb
+
+def is_bwd_hdim_supported(d):
+    return d <= 256
+
+
+def ck_randval_to_dropout_mask(randval, p):
+    # If p = 0.3, randval in 255 * (0.7, 1.0] will be dropout
+    # randval in 255 * [0, 0.7] will be kept
+    # If return dropout_mask >=0, value will be kept
+    return math.floor(255.0 * (1 - p)) - randval.to(torch.float32)
+
+
+def pad_rearrange_dropout_mask_hts_to_bhss(S_dmask, cu_seqlens_q, seqlen_q_rounded, seqlen_k_rounded):
+    """ pad + rearrange [nheads, total_q, max_seqlen_k] into [b, nheads, seqlen_q_rounded, seqlen_k_rounded]
+    Arguments:
+        S_dmask: (nheads, total_q, max_seqlen_k)
+        cu_seqlens_q: (b + 1)
+    Output:
+        S_dmask: (b, nheads, seqlen_q_rounded, seqlen_k_rounded)
+    """
+    batch_size = cu_seqlens_q.numel() - 1
+    seqlens_q = torch.roll(cu_seqlens_q, shifts = -1) - cu_seqlens_q
+    seqlens_q = seqlens_q[0:batch_size].tolist()
+    S_dmask = torch.split(S_dmask, seqlens_q, dim=1)
+    # [(nheads, seqlen_q0, max_seqlen_k), (nheads, seqlen_q1, max_seqlen_k), ..., (nheads, seqlen_qb, max_seqlen_k)]
+    masks = ()
+    for mask in S_dmask:
+        # (nheads, seqlen_qi, max_seqlen_k) -> (nheads, seqlen_q_rounded, seqlen_k_rounded)
+        mask = F.pad(mask, (0, seqlen_k_rounded - mask.shape[2], 0, seqlen_q_rounded - mask.shape[1], 0, 0)).unsqueeze(1)
+        masks = masks + (mask, )
+    S_dmask = torch.cat(masks, dim=1)
+
+    S_dmask = S_dmask.transpose(0, 1)
+    return S_dmask
+
+
+@pytest.mark.parametrize("dtype", [torch.float16, torch.bfloat16])
+@pytest.mark.parametrize("deterministic", [False, True])
+@pytest.mark.parametrize("alibi", [False, True])
+@pytest.mark.parametrize("local", [False, True])
+@pytest.mark.parametrize("causal", [False, True])
+@pytest.mark.parametrize("d", [32, 40, 59, 64, 80, 96, 111, 128, 160, 192, 224, 256])
+@pytest.mark.parametrize("seqlen", [97, 128, 200, 384, 768, 1024, 1025, 2048])
+@pytest.mark.parametrize("dropout_p", [0.0, 0.17])
+def test_flash_attn_qkvpacked(seqlen, d, dropout_p, causal, local, alibi, deterministic, dtype):
+    if d > 256:
+        pytest.skip()
+
+    device = "cuda"
+    # set seed
+    torch.random.manual_seed(0)
+    batch_size = 4
+    nheads = 9
+    window_size = (-1, -1) if not local else torch.randint(0, seqlen, (2,))
+
+    qkv = torch.randn(
+        batch_size, seqlen, 3, nheads, d, device=device, dtype=dtype, requires_grad=True
+    )
+
+    if alibi:
+        alibi_slopes = torch.rand(batch_size, nheads, device=device, dtype=torch.float32) * 0.3
+        attn_bias = attn_bias_from_alibi_slopes(alibi_slopes, seqlen, seqlen, causal=causal)
+    else:
+        alibi_slopes, attn_bias = None, None
+    out, lse, S_dmask = flash_attn_qkvpacked_func(
+        qkv,
+        dropout_p,
+        causal=causal,
+        window_size=window_size,
+        alibi_slopes=alibi_slopes,
+        deterministic=deterministic,
+        return_attn_probs=True,
+    )
+    if dropout_p > 0.0:
+        # TODO - move to c++ mha_varlen_fwd()
+        S_dmask = ck_randval_to_dropout_mask(S_dmask, dropout_p)
+        S_dmask_converted = convert_flash_attn_S_to_softmax(
+            S_dmask,
+            seqlen,
+            seqlen,
+            None,
+            None,
+            d,
+            dropout_p > 0.0,
+            causal=causal,
+            window_size=window_size,
+        )
+        dropout_mask = S_dmask_converted >= 0
+        # CK does not return P. Hence, we don't test the attn here.
+    else:
+        dropout_mask = None
+
+    out_ref, attn_ref = attention_qkvpacked_ref(
+        qkv, None, attn_bias, dropout_p, dropout_mask, causal=causal, window_size=window_size
+    )
+    out_pt, attn_pt = attention_qkvpacked_ref(
+        qkv,
+        None,
+        attn_bias,
+        dropout_p,
+        dropout_mask,
+        causal=causal,
+        window_size=window_size,
+        upcast=False,
+        reorder_ops=True,
+    )
+
+    print(f"Output max diff: {(out - out_ref).abs().max().item()}")
+    print(f"Output mean diff: {(out - out_ref).abs().mean().item()}")
+    print(f"Pytorch max diff: {(out_pt - out_ref).abs().max().item()}")
+    print(f"Pytorch mean diff: {(out_pt - out_ref).abs().mean().item()}")
+
+    # Check that FlashAttention's numerical error is at most twice the numerical error
+    # of a Pytorch implementation.
+    assert (out - out_ref).abs().max().item() <= 2 * (out_pt - out_ref).abs().max().item()
+
+    g = torch.randn_like(out)
+    if is_bwd_hdim_supported(d):
+        (dqkv,) = torch.autograd.grad(out, qkv, g)
+        (dqkv_ref,) = torch.autograd.grad(out_ref, qkv, g)
+        (dqkv_pt,) = torch.autograd.grad(out_pt, qkv, g)
+        print(f"dQ max diff: {(dqkv[:, :, 0] - dqkv_ref[:, :, 0]).abs().max().item()}")
+        print(f"dK max diff: {(dqkv[:, :, 1] - dqkv_ref[:, :, 1]).abs().max().item()}")
+        print(f"dV max diff: {(dqkv[:, :, 2] - dqkv_ref[:, :, 2]).abs().max().item()}")
+        print(f"dQKV mean diff: {(dqkv - dqkv_ref).abs().mean().item()}")
+        print(f"dQ Pytorch max diff: {(dqkv_pt[:, :, 0] - dqkv_ref[:, :, 0]).abs().max().item()}")
+        print(f"dK Pytorch max diff: {(dqkv_pt[:, :, 1] - dqkv_ref[:, :, 1]).abs().max().item()}")
+        print(f"dV Pytorch max diff: {(dqkv_pt[:, :, 2] - dqkv_ref[:, :, 2]).abs().max().item()}")
+        print(f"dQKV Pytorch mean diff: {(dqkv_pt - dqkv_ref).abs().mean().item()}")
+
+        # TODO - use 10 times to check, wait for ck to fix bwd precision issue
+        assert (dqkv - dqkv_ref).abs().max().item() <= 10 * (dqkv_pt - dqkv_ref).abs().max().item()
+
+
+@pytest.mark.parametrize("dtype", [torch.float16, torch.bfloat16])
+@pytest.mark.parametrize("deterministic", [False, True])
+@pytest.mark.parametrize("alibi", [False, True])
+@pytest.mark.parametrize("local", [False, True])
+@pytest.mark.parametrize("causal", [False, True])
+@pytest.mark.parametrize("d", [32, 59, 64, 80, 96, 128, 160, 192, 224, 256])
+@pytest.mark.parametrize("seqlen", [97, 128, 200, 257, 384, 512, 768, 1025, 2048])
+@pytest.mark.parametrize("dropout_p", [0, 0.17])
+def test_flash_attn_varlen_qkvpacked(seqlen, d, dropout_p, causal, local, alibi, deterministic, dtype):
+    if d > 256:
+        pytest.skip()
+
+    device = "cuda"
+    # set seed
+    torch.random.manual_seed(0)
+    batch_size = 5
+    nheads = 6
+    window_size = (-1, -1) if not local else torch.randint(0, seqlen, (2,))
+    qkv = torch.randn(
+        batch_size, seqlen, 3, nheads, d, device=device, dtype=dtype, requires_grad=True
+    )
+
+    key_padding_mask = generate_random_padding_mask(seqlen, batch_size, device, mode="random")
+    # key_padding_mask = generate_random_padding_mask(seqlen, batch_size, device, mode='full')
+    if alibi:
+        alibi_slopes = torch.rand(batch_size, nheads, device=device, dtype=torch.float32) * 0.3
+        attn_bias = attn_bias_from_alibi_slopes(
+            alibi_slopes, seqlen, seqlen, key_padding_mask, key_padding_mask, causal=causal
+        )
+    else:
+        alibi_slopes, attn_bias = None, None
+
+    qkv_unpad, cu_seqlens, max_seqlen, qkv, output_pad_fn, dqkv_pad_fn = generate_qkv(
+        *qkv.unbind(dim=2), key_padding_mask, key_padding_mask, qkvpacked=True
+    )
+
+    out_unpad, sm_lse, S_dmask = flash_attn_varlen_qkvpacked_func(
+        qkv_unpad,
+        cu_seqlens,
+        max_seqlen,
+        dropout_p,
+        causal=causal,
+        window_size=window_size,
+        alibi_slopes=alibi_slopes,
+        deterministic=deterministic,
+        return_attn_probs=True,
+    )
+    out = output_pad_fn(out_unpad)
+    if dropout_p > 0.0:
+        # TODO - move to c++ mha_varlen_fwd()
+        S_dmask = ck_randval_to_dropout_mask(S_dmask, dropout_p)
+        S_dmask = pad_rearrange_dropout_mask_hts_to_bhss(S_dmask, cu_seqlens, seqlen, seqlen)
+
+        S_dmask_converted = convert_flash_attn_S_to_softmax(
+            S_dmask,
+            seqlen,
+            seqlen,
+            key_padding_mask,
+            key_padding_mask,
+            d,
+            dropout_p > 0.0,
+            causal=causal,
+            window_size=window_size,
+        )
+
+        dropout_mask = S_dmask_converted >= 0
+        # CK does not return P. Hence, we don't test the attn here.
+    else:
+        dropout_mask = None
+
+    out_ref, attn_ref = attention_qkvpacked_ref(
+        qkv,
+        key_padding_mask,
+        attn_bias,
+        dropout_p,
+        dropout_mask,
+        causal=causal,
+        window_size=window_size,
+    )
+    out_pt, attn_pt = attention_qkvpacked_ref(
+        qkv,
+        key_padding_mask,
+        attn_bias,
+        dropout_p,
+        dropout_mask,
+        causal=causal,
+        window_size=window_size,
+        upcast=False,
+        reorder_ops=True,
+    )
+    print(f"Output max diff: {(out - out_ref).abs().max().item()}")
+    print(f"Output mean diff: {(out - out_ref).abs().mean().item()}")
+    print(f"Pytorch max diff: {(out_pt - out_ref).abs().max().item()}")
+    print(f"Pytorch mean diff: {(out_pt - out_ref).abs().mean().item()}")
+
+    # Check that FlashAttention's numerical error is at most twice the numerical error
+    # of a Pytorch implementation.
+    assert (out - out_ref).abs().max().item() <= 2 * (out_pt - out_ref).abs().max().item()
+
+    g = torch.randn_like(out)
+    if is_bwd_hdim_supported(d):
+        (dqkv_unpad,) = torch.autograd.grad(out, qkv_unpad, g)
+        dqkv = dqkv_pad_fn(dqkv_unpad)
+        (dqkv_ref,) = torch.autograd.grad(out_ref, qkv, g)
+        (dqkv_pt,) = torch.autograd.grad(out_pt, qkv, g)
+        print(f"dQ max diff: {(dqkv[:, :, 0] - dqkv_ref[:, :, 0]).abs().max().item()}")
+        print(f"dK max diff: {(dqkv[:, :, 1] - dqkv_ref[:, :, 1]).abs().max().item()}")
+        print(f"dV max diff: {(dqkv[:, :, 2] - dqkv_ref[:, :, 2]).abs().max().item()}")
+        print(f"dQKV mean diff: {(dqkv - dqkv_ref).abs().mean().item()}")
+        print(f"dQ Pytorch max diff: {(dqkv_pt[:, :, 0] - dqkv_ref[:, :, 0]).abs().max().item()}")
+        print(f"dK Pytorch max diff: {(dqkv_pt[:, :, 1] - dqkv_ref[:, :, 1]).abs().max().item()}")
+        print(f"dV Pytorch max diff: {(dqkv_pt[:, :, 2] - dqkv_ref[:, :, 2]).abs().max().item()}")
+        print(f"dQKV Pytorch mean diff: {(dqkv_pt - dqkv_ref).abs().mean().item()}")
+
+        # TODO - use 10 times to check, wait for ck to fix bwd precision issue
+        assert (dqkv - dqkv_ref).abs().max().item() <= 10 * (dqkv_pt - dqkv_ref).abs().max().item()
+
+
+@pytest.mark.parametrize("kvpacked", [True, False])
+@pytest.mark.parametrize("dtype", [torch.float16, torch.bfloat16])
+@pytest.mark.parametrize("mha_type", ["mha", "mqa", "gqa"])
+@pytest.mark.parametrize("deterministic", [False, True])
+@pytest.mark.parametrize("alibi", [False, True])
+@pytest.mark.parametrize("local", [False, True])
+@pytest.mark.parametrize("causal", [False, True])
+@pytest.mark.parametrize("d", [32, 40, 59, 64, 96, 111, 128, 160, 192, 224, 256])
+@pytest.mark.parametrize(
+    "seqlen_q,seqlen_k",
+    [
+        (113, 203),
+        (128, 217),
+        (113, 211),
+        (108, 256),
+        (256, 512),
+        (512, 256),
+        (1024, 1024),
+        (1023, 1024),
+        (1024, 1023),
+        (2048, 2048),
+    ],
+)
+@pytest.mark.parametrize("dropout_p", [0.0, 0.17])
+def test_flash_attn_output(
+    seqlen_q, seqlen_k, d, dropout_p, causal, local, alibi, deterministic, mha_type, dtype, kvpacked
+):
+    device = "cuda"
+    # set seed
+    torch.random.manual_seed(0)
+    batch_size = 4
+    nheads = 9
+    nheads_k = nheads if mha_type == "mha" else (1 if mha_type == "mqa" else 3)
+    assert nheads % nheads_k == 0
+    window_size = (-1, -1) if not local else torch.randint(0, seqlen_k, (2,))
+    q = torch.randn(batch_size, seqlen_q, nheads, d, device=device, dtype=dtype, requires_grad=True)
+    if kvpacked:
+        kv = torch.randn(
+            batch_size, seqlen_k, 2, nheads_k, d, device=device, dtype=dtype, requires_grad=True
+        )
+    else:
+        k = torch.randn(
+            batch_size, seqlen_k, nheads_k, d, device=device, dtype=dtype, requires_grad=True
+        )
+        v = torch.randn(
+            batch_size, seqlen_k, nheads_k, d, device=device, dtype=dtype, requires_grad=True
+        )
+    if alibi:
+        alibi_slopes = torch.rand(batch_size, nheads, device=device, dtype=torch.float32) * 0.3
+        attn_bias = attn_bias_from_alibi_slopes(alibi_slopes, seqlen_q, seqlen_k, causal=causal)
+    else:
+        alibi_slopes, attn_bias = None, None
+
+    if kvpacked:
+        out, lse, S_dmask = flash_attn_kvpacked_func(
+            q,
+            kv,
+            dropout_p,
+            causal=causal,
+            window_size=window_size,
+            alibi_slopes=alibi_slopes,
+            deterministic=deterministic,
+            return_attn_probs=True,
+        )
+    else:
+        out, lse, S_dmask = flash_attn_func(
+            q,
+            k,
+            v,
+            dropout_p,
+            causal=causal,
+            window_size=window_size,
+            alibi_slopes=alibi_slopes,
+            deterministic=deterministic,
+            return_attn_probs=True,
+        )
+    if dropout_p > 0.0:
+        # TODO - move to c++ mha_varlen_fwd()
+        S_dmask = ck_randval_to_dropout_mask(S_dmask, dropout_p)
+        S_dmask_converted = convert_flash_attn_S_to_softmax(
+            S_dmask,
+            seqlen_q,
+            seqlen_k,
+            None,
+            None,
+            d,
+            dropout_p > 0.0,
+            causal=causal,
+            window_size=window_size,
+        )
+        dropout_mask = S_dmask_converted >= 0
+        if kvpacked:
+            kv_rep = repeat(kv, "b s two h d -> b s two (h g) d", g=nheads // nheads_k)
+            k_rep, v_rep = kv_rep.unbind(dim=2)
+        else:
+            k_rep = repeat(k, "b s h d -> b s (h g) d", g=nheads // nheads_k)
+            v_rep = repeat(v, "b s h d -> b s (h g) d", g=nheads // nheads_k)
+        # CK does not return P. Hence, we don't test the attn here.
+    else:
+        dropout_mask = None
+
+    if kvpacked:
+        out_ref, attn_ref = attention_kvpacked_ref(
+            q,
+            kv,
+            None,
+            None,
+            attn_bias,
+            dropout_p,
+            dropout_mask,
+            causal=causal,
+            window_size=window_size,
+        )
+        out_pt, attn_pt = attention_kvpacked_ref(
+            q,
+            kv,
+            None,
+            None,
+            attn_bias,
+            dropout_p,
+            dropout_mask,
+            causal=causal,
+            window_size=window_size,
+            upcast=False,
+            reorder_ops=True,
+        )
+    else:
+        out_ref, attn_ref = attention_ref(
+            q,
+            k,
+            v,
+            None,
+            None,
+            attn_bias,
+            dropout_p,
+            dropout_mask,
+            causal=causal,
+            window_size=window_size,
+        )
+        out_pt, attn_pt = attention_ref(
+            q,
+            k,
+            v,
+            None,
+            None,
+            attn_bias,
+            dropout_p,
+            dropout_mask,
+            causal=causal,
+            window_size=window_size,
+            upcast=False,
+            reorder_ops=True,
+        )
+
+    print(f"Output max diff: {(out - out_ref).abs().max().item()}")
+    print(f"Output mean diff: {(out - out_ref).abs().mean().item()}")
+    print(f"Pytorch max diff: {(out_pt - out_ref).abs().max().item()}")
+    print(f"Pytorch mean diff: {(out_pt - out_ref).abs().mean().item()}")
+
+    # Check that FlashAttention's numerical error is at most twice the numerical error
+    # of a Pytorch implementation.
+    assert (out - out_ref).abs().max().item() <= 2 * (out_pt - out_ref).abs().max().item()
+
+    g = torch.randn_like(out)
+    if is_bwd_hdim_supported(d):
+        if kvpacked:
+            (
+                dq,
+                dkv,
+            ) = torch.autograd.grad(out, (q, kv), g)
+            dk, dv = dkv.unbind(2)
+            (
+                dq_ref,
+                dkv_ref,
+            ) = torch.autograd.grad(out_ref, (q, kv), g)
+            dk_ref, dv_ref = dkv_ref.unbind(2)
+            (
+                dq_pt,
+                dkv_pt,
+            ) = torch.autograd.grad(out_pt, (q, kv), g)
+            dk_pt, dv_pt = dkv_pt.unbind(2)
+        else:
+            (
+                dq,
+                dk,
+                dv,
+            ) = torch.autograd.grad(out, (q, k, v), g)
+            (
+                dq_ref,
+                dk_ref,
+                dv_ref,
+            ) = torch.autograd.grad(out_ref, (q, k, v), g)
+            (
+                dq_pt,
+                dk_pt,
+                dv_pt,
+            ) = torch.autograd.grad(out_pt, (q, k, v), g)
+        print(f"dQ max diff: {(dq - dq_ref).abs().max().item()}")
+        print(f"dK max diff: {(dk - dk_ref).abs().max().item()}")
+        print(f"dV max diff: {(dv - dv_ref).abs().max().item()}")
+        print(f"dQ mean diff: {(dq - dq_ref).abs().mean().item()}")
+        print(f"dK mean diff: {(dk - dk_ref).abs().mean().item()}")
+        print(f"dV mean diff: {(dv - dv_ref).abs().mean().item()}")
+        print(f"dQ Pytorch max diff: {(dq_pt - dq_ref).abs().max().item()}")
+        print(f"dK Pytorch max diff: {(dk_pt - dk_ref).abs().max().item()}")
+        print(f"dV Pytorch max diff: {(dv_pt - dv_ref).abs().max().item()}")
+        print(f"dQ Pytorch mean diff: {(dq_pt - dq_ref).abs().mean().item()}")
+        print(f"dK Pytorch mean diff: {(dk_pt - dk_ref).abs().mean().item()}")
+        print(f"dV Pytorch mean diff: {(dv_pt - dv_ref).abs().mean().item()}")
+
+        # TODO - use 10 times to check, wait for ck to fix bwd precision issue
+        assert (dq - dq_ref).abs().max().item() <= 10 * (dq_pt - dq_ref).abs().max().item()
+        assert (dk - dk_ref).abs().max().item() <= 10 * (dk_pt - dk_ref).abs().max().item()
+        assert (dv - dv_ref).abs().max().item() <= 10 * (dv_pt - dv_ref).abs().max().item()
+
+
+@pytest.mark.parametrize("kvpacked", [True, False])
+@pytest.mark.parametrize("dtype", [torch.float16, torch.bfloat16])
+@pytest.mark.parametrize("mha_type", ["mha", "mqa", "gqa"])
+@pytest.mark.parametrize("deterministic", [False, True])
+@pytest.mark.parametrize("alibi", [False, True])
+@pytest.mark.parametrize("local", [False, True])
+@pytest.mark.parametrize("causal", [False, True])
+@pytest.mark.parametrize("d", [32, 59, 64, 80, 96, 111, 128, 160, 192, 224, 256])
+@pytest.mark.parametrize(
+    "seqlen_q,seqlen_k",
+    [
+        (1, 147),
+        (113, 203),
+        (128, 217),
+        (113, 211),
+        (108, 256),
+        (256, 512),
+        (512, 256),
+        (1024, 1024),
+        (1023, 1024),
+        (1024, 1023),
+        (2048, 2048),
+    ],
+)
+@pytest.mark.parametrize("dropout_p", [0.0, 0.17])
+def test_flash_attn_varlen_output(
+    seqlen_q, seqlen_k, d, dropout_p, causal, local, alibi, deterministic, mha_type, dtype, kvpacked
+):
+    device = "cuda"
+    # set seed
+    torch.random.manual_seed(0)
+    batch_size = 4
+    nheads = 9
+    nheads_k = nheads if mha_type == "mha" else (1 if mha_type == "mqa" else 3)
+    assert nheads % nheads_k == 0
+    window_size = (-1, -1) if not local else torch.randint(0, seqlen_k, (2,))
+    q = torch.randn(batch_size, seqlen_q, nheads, d, device=device, dtype=dtype, requires_grad=True)
+    if kvpacked:
+        kv = torch.randn(
+            batch_size, seqlen_k, 2, nheads_k, d, device=device, dtype=dtype, requires_grad=True
+        )
+    else:
+        k = torch.randn(
+            batch_size, seqlen_k, nheads_k, d, device=device, dtype=dtype, requires_grad=True
+        )
+        v = torch.randn(
+            batch_size, seqlen_k, nheads_k, d, device=device, dtype=dtype, requires_grad=True
+        )
+
+    query_padding_mask = generate_random_padding_mask(seqlen_q, batch_size, device, mode="random")
+    key_padding_mask = generate_random_padding_mask(seqlen_k, batch_size, device, mode="random")
+    # key_padding_mask = generate_random_padding_mask(seqlen_k, batch_size, device, mode='full')
+    if alibi:
+        alibi_slopes = torch.rand(batch_size, nheads, device=device, dtype=torch.float32) * 0.3
+        attn_bias = attn_bias_from_alibi_slopes(
+            alibi_slopes, seqlen_q, seqlen_k, query_padding_mask, key_padding_mask, causal=causal
+        )
+    else:
+        alibi_slopes, attn_bias = None, None
+
+    if kvpacked:
+        (
+            q_unpad,
+            kv_unpad,
+            cu_seqlens_q,
+            cu_seqlens_k,
+            max_seqlen_q,
+            max_seqlen_k,
+            q,
+            kv,
+            output_pad_fn,
+            dq_pad_fn,
+            dkv_pad_fn,
+        ) = generate_qkv(q, *kv.unbind(dim=2), query_padding_mask, key_padding_mask, kvpacked=True)
+        out_unpad, sm_lse, S_dmask = flash_attn_varlen_kvpacked_func(
+            q_unpad,
+            kv_unpad,
+            cu_seqlens_q,
+            cu_seqlens_k,
+            max_seqlen_q,
+            max_seqlen_k,
+            dropout_p,
+            causal=causal,
+            window_size=window_size,
+            alibi_slopes=alibi_slopes,
+            deterministic=deterministic,
+            return_attn_probs=True,
+        )
+    else:
+        (
+            q_unpad,
+            k_unpad,
+            v_unpad,
+            cu_seqlens_q,
+            cu_seqlens_k,
+            max_seqlen_q,
+            max_seqlen_k,
+            q,
+            k,
+            v,
+            output_pad_fn,
+            dq_pad_fn,
+            dk_pad_fn,
+        ) = generate_qkv(q, k, v, query_padding_mask, key_padding_mask, kvpacked=False)
+        out_unpad, sm_lse, S_dmask = flash_attn_varlen_func(
+            q_unpad,
+            k_unpad,
+            v_unpad,
+            cu_seqlens_q,
+            cu_seqlens_k,
+            max_seqlen_q,
+            max_seqlen_k,
+            dropout_p,
+            causal=causal,
+            window_size=window_size,
+            alibi_slopes=alibi_slopes,
+            deterministic=deterministic,
+            return_attn_probs=True,
+        )
+    out = output_pad_fn(out_unpad)
+    if dropout_p > 0.0:
+        # TODO - move to c++ mha_varlen_fwd()
+        S_dmask = ck_randval_to_dropout_mask(S_dmask, dropout_p)
+        S_dmask = pad_rearrange_dropout_mask_hts_to_bhss(S_dmask, cu_seqlens_q, seqlen_q, seqlen_k)
+        S_dmask_converted = convert_flash_attn_S_to_softmax(
+            S_dmask,
+            seqlen_q,
+            seqlen_k,
+            query_padding_mask,
+            key_padding_mask,
+            d,
+            dropout_p > 0.0,
+            causal=causal,
+            window_size=window_size,
+        )
+        dropout_mask = S_dmask_converted >= 0
+        if kvpacked:
+            kv_rep = repeat(kv, "b s two h d -> b s two (h g) d", g=nheads // nheads_k)
+            k_rep, v_rep = kv_rep.unbind(dim=2)
+        else:
+            k_rep = repeat(k, "b s h d -> b s (h g) d", g=nheads // nheads_k)
+            v_rep = repeat(v, "b s h d -> b s (h g) d", g=nheads // nheads_k)
+        # CK does not return P. Hence, we don't test the attn here.
+    else:
+        dropout_mask = None
+
+    if kvpacked:
+        out_ref, attn_ref = attention_kvpacked_ref(
+            q,
+            kv,
+            query_padding_mask,
+            key_padding_mask,
+            attn_bias,
+            dropout_p,
+            dropout_mask,
+            causal=causal,
+            window_size=window_size,
+        )
+        out_pt, attn_pt = attention_kvpacked_ref(
+            q,
+            kv,
+            query_padding_mask,
+            key_padding_mask,
+            attn_bias,
+            dropout_p,
+            dropout_mask,
+            causal=causal,
+            window_size=window_size,
+            upcast=False,
+            reorder_ops=True,
+        )
+    else:
+        out_ref, attn_ref = attention_ref(
+            q,
+            k,
+            v,
+            query_padding_mask,
+            key_padding_mask,
+            attn_bias,
+            dropout_p,
+            dropout_mask,
+            causal=causal,
+            window_size=window_size,
+        )
+        out_pt, attn_pt = attention_ref(
+            q,
+            k,
+            v,
+            query_padding_mask,
+            key_padding_mask,
+            attn_bias,
+            dropout_p,
+            dropout_mask,
+            causal=causal,
+            window_size=window_size,
+            upcast=False,
+            reorder_ops=True,
+        )
+
+    print(f"Output max diff: {(out - out_ref).abs().max().item()}")
+    print(f"Output mean diff: {(out - out_ref).abs().mean().item()}")
+    print(f"Pytorch max diff: {(out_pt - out_ref).abs().max().item()}")
+    print(f"Pytorch mean diff: {(out_pt - out_ref).abs().mean().item()}")
+
+    # Check that FlashAttention's numerical error is at most 4 times the numerical error
+    # of a Pytorch implementation.
+    assert (out - out_ref).abs().max().item() <= 4 * (out_pt - out_ref).abs().max().item()
+
+    g = torch.randn_like(out)
+    if is_bwd_hdim_supported(d):
+        if kvpacked:
+            (
+                dq_unpad,
+                dkv_unpad,
+            ) = torch.autograd.grad(out, (q_unpad, kv_unpad), g)
+            dk, dv = dkv_pad_fn(dkv_unpad).unbind(2)
+            (
+                dq_ref,
+                dkv_ref,
+            ) = torch.autograd.grad(out_ref, (q, kv), g)
+            dk_ref, dv_ref = dkv_ref.unbind(2)
+            (
+                dq_pt,
+                dkv_pt,
+            ) = torch.autograd.grad(out_pt, (q, kv), g)
+            dk_pt, dv_pt = dkv_pt.unbind(2)
+        else:
+            (
+                dq_unpad,
+                dk_unpad,
+                dv_unpad,
+            ) = torch.autograd.grad(out, (q_unpad, k_unpad, v_unpad), g)
+            dk = dk_pad_fn(dk_unpad)
+            dv = dk_pad_fn(dv_unpad)
+            (
+                dq_ref,
+                dk_ref,
+                dv_ref,
+            ) = torch.autograd.grad(out_ref, (q, k, v), g)
+            (
+                dq_pt,
+                dk_pt,
+                dv_pt,
+            ) = torch.autograd.grad(out_pt, (q, k, v), g)
+        dq = dq_pad_fn(dq_unpad)
+        print(f"dQ max diff: {(dq - dq_ref).abs().max().item()}")
+        print(f"dK max diff: {(dk - dk_ref).abs().max().item()}")
+        print(f"dV max diff: {(dv - dv_ref).abs().max().item()}")
+        print(f"dQ mean diff: {(dq - dq_ref).abs().mean().item()}")
+        print(f"dK mean diff: {(dk - dk_ref).abs().mean().item()}")
+        print(f"dV mean diff: {(dv - dv_ref).abs().mean().item()}")
+        print(f"dQ Pytorch max diff: {(dq_pt - dq_ref).abs().max().item()}")
+        print(f"dK Pytorch max diff: {(dk_pt - dk_ref).abs().max().item()}")
+        print(f"dV Pytorch max diff: {(dv_pt - dv_ref).abs().max().item()}")
+        print(f"dQ Pytorch mean diff: {(dq_pt - dq_ref).abs().mean().item()}")
+        print(f"dK Pytorch mean diff: {(dk_pt - dk_ref).abs().mean().item()}")
+        print(f"dV Pytorch mean diff: {(dv_pt - dv_ref).abs().mean().item()}")
+
+        # TODO - use 10 times to check, wait for ck to fix bwd precision issue
+        assert (dq - dq_ref).abs().max().item() <= 10 * (dq_pt - dq_ref).abs().max().item()
+        assert (dk - dk_ref).abs().max().item() <= 10 * (dk_pt - dk_ref).abs().max().item()
+        assert (dv - dv_ref).abs().max().item() <= 10 * (dv_pt - dv_ref).abs().max().item()
+
+
+@pytest.mark.parametrize("dtype", [torch.float16, torch.bfloat16])
+@pytest.mark.parametrize("local", [False, True])
+@pytest.mark.parametrize("d", [32, 40, 59, 64, 80, 96, 111, 128, 160, 192, 224, 256])
+@pytest.mark.parametrize("swap_sq_sk", [False, True])
+@pytest.mark.parametrize(
+    "seqlen_q,seqlen_k",
+    [
+        # (1, 239),
+        (3, 799),
+        (127, 512),
+        (127, 513),
+        (113, 203),
+        (128, 217),
+        (113, 211),
+        (108, 256),
+        (256, 512),
+        (1023, 1024),
+    ],
+)
+def test_flash_attn_causal(seqlen_q, seqlen_k, swap_sq_sk, d, local, dtype):
+    if max(seqlen_q, seqlen_k) >= 2048:
+        pytest.skip()
+    if swap_sq_sk:
+        seqlen_q, seqlen_k = seqlen_k, seqlen_q
+    device = "cuda"
+    causal = True
+    # set seed
+    torch.random.manual_seed(0)
+    batch_size = 8
+    nheads = 9
+    window_size = (-1, -1) if not local else torch.randint(0, seqlen_k, (2,))
+    q = torch.randn(batch_size, seqlen_q, nheads, d, device=device, dtype=dtype, requires_grad=True)
+    k = torch.randn(batch_size, seqlen_k, nheads, d, device=device, dtype=dtype, requires_grad=True)
+    v = torch.randn(batch_size, seqlen_k, nheads, d, device=device, dtype=dtype, requires_grad=True)
+    out = flash_attn_func(q, k, v, 0.0, causal=causal, window_size=window_size)
+    out_ref, attn_ref = attention_ref(
+        q, k, v, None, None, None, 0.0, None, causal=causal, window_size=window_size
+    )
+    out_pt, attn_pt = attention_ref(
+        q,
+        k,
+        v,
+        None,
+        None,
+        None,
+        0.0,
+        None,
+        causal=causal,
+        window_size=window_size,
+        upcast=False,
+        reorder_ops=True,
+    )
+
+    print(f"Output max diff: {(out - out_ref).abs().max().item()}")
+    print(f"Output mean diff: {(out - out_ref).abs().mean().item()}")
+    print(f"Pytorch max diff: {(out_pt - out_ref).abs().max().item()}")
+    print(f"Pytorch mean diff: {(out_pt - out_ref).abs().mean().item()}")
+
+    # Check that FlashAttention's numerical error is at most 4 times the numerical error
+    # of a Pytorch implementation.
+    assert (out - out_ref).abs().max().item() <= 4 * (out_pt - out_ref).abs().max().item() + 1e-5
+
+    g = torch.randn_like(out)
+    if is_bwd_hdim_supported(d):
+        do_o = (g.float() * out.float()).sum(-1)
+        (
+            dq,
+            dk,
+            dv,
+        ) = torch.autograd.grad(out, (q, k, v), g)
+        (
+            dq_ref,
+            dk_ref,
+            dv_ref,
+        ) = torch.autograd.grad(out_ref, (q, k, v), g)
+        (
+            dq_pt,
+            dk_pt,
+            dv_pt,
+        ) = torch.autograd.grad(out_pt, (q, k, v), g)
+        print(f"dQ max diff: {(dq - dq_ref).abs().max().item()}")
+        print(f"dK max diff: {(dk - dk_ref).abs().max().item()}")
+        print(f"dV max diff: {(dv - dv_ref).abs().max().item()}")
+        print(f"dQ mean diff: {(dq - dq_ref).abs().mean().item()}")
+        print(f"dK mean diff: {(dk - dk_ref).abs().mean().item()}")
+        print(f"dV mean diff: {(dv - dv_ref).abs().mean().item()}")
+        print(f"dQ Pytorch max diff: {(dq_pt - dq_ref).abs().max().item()}")
+        print(f"dK Pytorch max diff: {(dk_pt - dk_ref).abs().max().item()}")
+        print(f"dV Pytorch max diff: {(dv_pt - dv_ref).abs().max().item()}")
+        print(f"dQ Pytorch mean diff: {(dq_pt - dq_ref).abs().mean().item()}")
+        print(f"dK Pytorch mean diff: {(dk_pt - dk_ref).abs().mean().item()}")
+        print(f"dV Pytorch mean diff: {(dv_pt - dv_ref).abs().mean().item()}")
+
+    # TODO - use 10 times to check, wait for ck to fix bwd precision issue
+    assert (dq - dq_ref).abs().max().item() <= 10 * (dq_pt - dq_ref).abs().max().item() + 1e-4
+    assert (dk - dk_ref).abs().max().item() <= 10 * (dk_pt - dk_ref).abs().max().item() + 1e-4
+    assert (dv - dv_ref).abs().max().item() <= 10 * (dv_pt - dv_ref).abs().max().item() + 1e-4
+
+
+@pytest.mark.parametrize("dtype", [torch.float16, torch.bfloat16])
+@pytest.mark.parametrize("local", [False, True])
+@pytest.mark.parametrize("d", [32, 40, 59, 64, 80, 96, 111, 128, 160, 192, 224, 256])
+@pytest.mark.parametrize("swap_sq_sk", [False, True])
+@pytest.mark.parametrize(
+    "seqlen_q,seqlen_k",
+    [
+        # (1, 239),
+        (3, 799),
+        (127, 512),
+        (127, 513),
+        (113, 203),
+        (128, 217),
+        (113, 211),
+        (108, 256),
+        (256, 512),
+        (1023, 1024),
+    ],
+)
+@pytest.mark.parametrize("paged_kv_block_size", [None, 256, 512])
+def test_flash_attn_varlen_causal(
+    seqlen_q, seqlen_k, swap_sq_sk, d, local, paged_kv_block_size, dtype
+):
+    if max(seqlen_q, seqlen_k) >= 2048:
+        pytest.skip()
+    if swap_sq_sk:
+        seqlen_q, seqlen_k = seqlen_k, seqlen_q
+    device = "cuda"
+    causal = True
+    # set seed
+    torch.random.manual_seed(0)
+    batch_size = 8
+    nheads = 9
+    window_size = (-1, -1) if not local else torch.randint(0, seqlen_k, (2,))
+    q = torch.randn(batch_size, seqlen_q, nheads, d, device=device, dtype=dtype, requires_grad=True)
+
+    if paged_kv_block_size is None:
+        k = torch.randn(
+            batch_size, seqlen_k, nheads, d, device=device, dtype=dtype, requires_grad=True
+        )
+        v = torch.randn(
+            batch_size, seqlen_k, nheads, d, device=device, dtype=dtype, requires_grad=True
+        )
+        block_table = None
+    else:
+        k, v, block_table, k_cache_paged, v_cache_paged, num_blocks = _generate_block_kvcache(
+            seqlen_k, paged_kv_block_size, batch_size, nheads, d, device, dtype
+        )
+    query_padding_mask = generate_random_padding_mask(seqlen_q, batch_size, device, mode="random")
+    key_padding_mask = generate_random_padding_mask(seqlen_k, batch_size, device, mode="random")
+    (
+        q_unpad,
+        k_unpad,
+        v_unpad,
+        cu_seqlens_q,
+        cu_seqlens_k,
+        max_seqlen_q,
+        max_seqlen_k,
+        q,
+        k,
+        v,
+        output_pad_fn,
+        dq_pad_fn,
+        dk_pad_fn,
+    ) = generate_qkv(q, k, v, query_padding_mask, key_padding_mask, kvpacked=False)
+    out_unpad = flash_attn_varlen_func(
+        q_unpad,
+        k_unpad if paged_kv_block_size is None else k_cache_paged,
+        v_unpad if paged_kv_block_size is None else v_cache_paged,
+        cu_seqlens_q,
+        cu_seqlens_k,
+        max_seqlen_q,
+        max_seqlen_k,
+        0.0,
+        causal=causal,
+        window_size=window_size,
+        block_table=block_table,
+    )
+    out = output_pad_fn(out_unpad)
+    out_ref, attn_ref = attention_ref(
+        q,
+        k,
+        v,
+        query_padding_mask,
+        key_padding_mask,
+        None,
+        0.0,
+        None,
+        causal=causal,
+        window_size=window_size,
+    )
+    out_pt, attn_pt = attention_ref(
+        q,
+        k,
+        v,
+        query_padding_mask,
+        key_padding_mask,
+        None,
+        0.0,
+        None,
+        causal=causal,
+        window_size=window_size,
+        upcast=False,
+        reorder_ops=True,
+    )
+
+    print(f"Output max diff: {(out - out_ref).abs().max().item()}")
+    print(f"Output mean diff: {(out - out_ref).abs().mean().item()}")
+    print(f"Pytorch max diff: {(out_pt - out_ref).abs().max().item()}")
+    print(f"Pytorch mean diff: {(out_pt - out_ref).abs().mean().item()}")
+
+    # Check that FlashAttention's numerical error is at most twice the numerical error
+    # of a Pytorch implementation.
+    assert (out - out_ref).abs().max().item() <= 2 * (out_pt - out_ref).abs().max().item() + 1e-5
+
+    g = torch.randn_like(out)
+    if is_bwd_hdim_supported(d):
+        do_o = (g.float() * out.float()).sum(-1)
+        test_backward = block_table is None
+        if test_backward:
+            (
+                dq_unpad,
+                dk_unpad,
+                dv_unpad,
+            ) = torch.autograd.grad(out, (q_unpad, k_unpad, v_unpad), g)
+            dq = dq_pad_fn(dq_unpad)
+            dk = dk_pad_fn(dk_unpad)
+            dv = dk_pad_fn(dv_unpad)
+            (
+                dq_ref,
+                dk_ref,
+                dv_ref,
+            ) = torch.autograd.grad(out_ref, (q, k, v), g)
+            (
+                dq_pt,
+                dk_pt,
+                dv_pt,
+            ) = torch.autograd.grad(out_pt, (q, k, v), g)
+            print(f"dQ max diff: {(dq - dq_ref).abs().max().item()}")
+            print(f"dK max diff: {(dk - dk_ref).abs().max().item()}")
+            print(f"dV max diff: {(dv - dv_ref).abs().max().item()}")
+            print(f"dQ mean diff: {(dq - dq_ref).abs().mean().item()}")
+            print(f"dK mean diff: {(dk - dk_ref).abs().mean().item()}")
+            print(f"dV mean diff: {(dv - dv_ref).abs().mean().item()}")
+            print(f"dQ Pytorch max diff: {(dq_pt - dq_ref).abs().max().item()}")
+            print(f"dK Pytorch max diff: {(dk_pt - dk_ref).abs().max().item()}")
+            print(f"dV Pytorch max diff: {(dv_pt - dv_ref).abs().max().item()}")
+            print(f"dQ Pytorch mean diff: {(dq_pt - dq_ref).abs().mean().item()}")
+            print(f"dK Pytorch mean diff: {(dk_pt - dk_ref).abs().mean().item()}")
+            print(f"dV Pytorch mean diff: {(dv_pt - dv_ref).abs().mean().item()}")
+
+        if test_backward:
+            # TODO - use 10 times to check, wait for ck to fix bwd precision issue
+            assert (dq - dq_ref).abs().max().item() <= 10 * (dq_pt - dq_ref).abs().max().item() + 1e-5
+            assert (dk - dk_ref).abs().max().item() <= 10 * (dk_pt - dk_ref).abs().max().item() + 1e-5
+            assert (dv - dv_ref).abs().max().item() <= 10 * (dv_pt - dv_ref).abs().max().item() + 1e-5
+
+
+# TODO - support splitkv
+# def test_flash_attn_splitkv
+
+
+# TODO - Support has_leftpad
+@pytest.mark.parametrize("dtype", [torch.float16])
+@pytest.mark.parametrize("num_splits", [1, 0])
+@pytest.mark.parametrize("mha_type", ["mha", "mqa", "gqa"])
+@pytest.mark.parametrize("new_kv", [False, True])
+@pytest.mark.parametrize("alibi", [False, True])
+@pytest.mark.parametrize("local", [False, True])
+@pytest.mark.parametrize("causal", [False, True])
+@pytest.mark.parametrize("seqlen_new_eq_seqlen_q", [True, False])
+@pytest.mark.parametrize("rotary_interleaved", [False, True])
+@pytest.mark.parametrize("rotary_fraction", [0.0, 0.5, 1.0])
+@pytest.mark.parametrize("paged_kv_block_size", [None, 256])
+@pytest.mark.parametrize("has_leftpad", [False])
+@pytest.mark.parametrize("has_batch_idx", [False, True])
+@pytest.mark.parametrize("d", [32, 59, 64, 80, 128, 256])
+@pytest.mark.parametrize(
+    "seqlen_q,seqlen_k",
+    [
+        (1, 128),
+        (1, 339),
+        (3, 1024),
+        (64, 800),
+        (64, 256),
+        (3, 799),
+        (64, 2048),
+        (16, 20000),
+        (1, 128 * 1024),
+        (16, 128 * 1024),
+        (128, 128),
+    ],
+)
+def test_flash_attn_kvcache(
+    seqlen_q,
+    seqlen_k,
+    d,
+    has_batch_idx,
+    has_leftpad,
+    paged_kv_block_size,
+    rotary_fraction,
+    rotary_interleaved,
+    seqlen_new_eq_seqlen_q,
+    causal,
+    local,
+    alibi,
+    new_kv,
+    mha_type,
+    num_splits,
+    dtype,
+):
+    if seqlen_q > seqlen_k and new_kv:
+        pytest.skip()
+    if not new_kv and rotary_fraction > 0.0:
+        pytest.skip()
+    if has_batch_idx and paged_kv_block_size is not None:
+        pytest.skip()
+    if has_leftpad and paged_kv_block_size is not None:
+        pytest.skip()
+    device = "cuda"
+    # set seed
+    torch.random.manual_seed(0)
+    batch_size = 1
+    batch_size_cache = batch_size if not has_batch_idx else batch_size * 2
+    nheads = 6
+    # rotary_dim must be a multiple of 16, and must be <= d
+    rotary_dim = math.floor(int(rotary_fraction * d) / 16) * 16
+    nheads_k = nheads if mha_type == "mha" else (1 if mha_type == "mqa" else 3)
+    assert nheads % nheads_k == 0
+    window_size = (-1, -1) if not local else torch.randint(0, seqlen_k, (2,))
+    q = torch.randn(batch_size, seqlen_q, nheads, d, device=device, dtype=dtype)
+    seqlen_new = seqlen_q if seqlen_new_eq_seqlen_q else torch.randint(1, seqlen_q + 1, (1,)).item()
+    if new_kv:
+        k = torch.randn(batch_size, seqlen_new, nheads_k, d, device=device, dtype=dtype)
+        v = torch.randn(batch_size, seqlen_new, nheads_k, d, device=device, dtype=dtype)
+    else:
+        k, v = None, None
+    if paged_kv_block_size is None:
+        k_cache = torch.randn(batch_size_cache, seqlen_k, nheads_k, d, device=device, dtype=dtype)
+        v_cache = torch.randn(batch_size_cache, seqlen_k, nheads_k, d, device=device, dtype=dtype)
+        block_table = None
+    else:
+        (
+            k_cache,
+            v_cache,
+            block_table,
+            k_cache_paged,
+            v_cache_paged,
+            num_blocks,
+        ) = _generate_block_kvcache(
+            seqlen_k, paged_kv_block_size, batch_size, nheads_k, d, device, dtype
+        )
+    cache_seqlens = torch.randint(
+        0 if new_kv else 1,
+        # If we don't use seqlen_q in the case of causal and rotary, cos/sin won't be long enough
+        (
+            (seqlen_k - (seqlen_q if (causal or local) and rotary_dim > 1 else seqlen_new) + 1)
+            if new_kv
+            else (seqlen_k + 1)
+        ),
+        (batch_size,),
+        dtype=torch.int32,
+        device=device,
+    )
+    if has_leftpad:
+        cache_leftpad = torch.cat([torch.randint(0, cache_seqlens[i].item(), (1,), dtype=torch.int32, device=device)
+                                   if cache_seqlens[i].item() > 0 else torch.zeros(1, dtype=torch.int32, device=device)
+                                   for i in range(batch_size)])
+    else:
+        cache_leftpad = None
+    arange = rearrange(torch.arange(seqlen_k, device=device), "s -> 1 s")
+    cache_seqlens_expanded = rearrange(cache_seqlens, "b -> b 1")
+    key_padding_mask = arange < cache_seqlens_expanded + (seqlen_new if new_kv else 0)
+    if has_leftpad:
+        key_padding_mask = torch.logical_and(
+            key_padding_mask, arange >= cache_leftpad.unsqueeze(-1).expand(-1, seqlen_k)
+        )
+    if has_batch_idx:
+        cache_batch_idx = torch.randperm(batch_size_cache, dtype=torch.int32, device=device)[
+            :batch_size
+        ]
+    else:
+        cache_batch_idx = None
+    if alibi:
+        alibi_slopes = torch.rand(batch_size, nheads, device=device, dtype=torch.float32) * 0.3
+        attn_bias = attn_bias_from_alibi_slopes(
+            alibi_slopes, seqlen_q, seqlen_k, None, key_padding_mask, causal=causal, key_leftpad=cache_leftpad
+        )
+    else:
+        alibi_slopes, attn_bias = None, None
+    # cache_seqlens = torch.tensor([64], dtype=torch.int32, device=device)
+    if rotary_dim > 0:
+        angle = (
+            torch.rand(
+                seqlen_k if paged_kv_block_size is None else num_blocks * paged_kv_block_size,
+                rotary_dim // 2,
+                device=device,
+            )
+            * 2
+            * math.pi
+        )
+        cos = torch.cos(angle).to(dtype=dtype)
+        sin = torch.sin(angle).to(dtype=dtype)
+        if causal or local:
+            q_ro = apply_rotary_emb(
+                q, cos, sin, seqlen_offsets=cache_seqlens, interleaved=rotary_interleaved
+            )
+        else:
+            q_ro = rearrange(
+                apply_rotary_emb(
+                    rearrange(q, "b s h d -> b 1 (s h) d"),
+                    cos,
+                    sin,
+                    seqlen_offsets=cache_seqlens,
+                    interleaved=rotary_interleaved,
+                ),
+                "b 1 (s h) d -> b s h d",
+                s=seqlen_q,
+            )
+        # q_ro = q
+        k_ro = apply_rotary_emb(
+            k, cos, sin, seqlen_offsets=cache_seqlens, interleaved=rotary_interleaved
+        )
+    else:
+        cos, sin = None, None
+        q_ro, k_ro = q, k
+    # k_cache[:, 64:] = -1
+    k_cache_ref = (
+        k_cache if not has_batch_idx else k_cache[cache_batch_idx.to(dtype=torch.long)]
+    ).clone()
+    v_cache_ref = (
+        v_cache if not has_batch_idx else v_cache[cache_batch_idx.to(dtype=torch.long)]
+    ).clone()
+    if new_kv:
+        update_mask = torch.logical_and(
+            cache_seqlens_expanded <= arange, arange < cache_seqlens_expanded + seqlen_new
+        )
+        k_cache_ref[update_mask] = rearrange(k_ro, "b s ... -> (b s) ...")
+        v_cache_ref[update_mask] = rearrange(v, "b s ... -> (b s) ...")
+    k_cache_rep = repeat(k_cache_ref, "b s h d -> b s (h g) d", g=nheads // nheads_k)
+    v_cache_rep = repeat(v_cache_ref, "b s h d -> b s (h g) d", g=nheads // nheads_k)
+    out = flash_attn_with_kvcache(
+        q,
+        k_cache if paged_kv_block_size is None else k_cache_paged,
+        v_cache if paged_kv_block_size is None else v_cache_paged,
+        k,
+        v,
+        rotary_cos=cos,
+        rotary_sin=sin,
+        cache_seqlens=cache_seqlens,
+        cache_batch_idx=cache_batch_idx,
+        cache_leftpad=cache_leftpad,
+        block_table=block_table,
+        causal=causal,
+        window_size=window_size,
+        rotary_interleaved=rotary_interleaved,
+        alibi_slopes=alibi_slopes,
+        num_splits=num_splits,
+    )
+    # out = flash_attn_with_kvcache(
+    #     q, k_cache, v_cache, cache_seqlens=cache_seqlens, causal=causal, window_size=window_size
+    # )
+    # out = flash_attn_with_kvcache(q, k_cache, v_cache, causal=causal, window_size=window_size)
+    # qk = torch.einsum("bqhd,bkhd->bhqk", q, k_cache_ref)
+    # m = qk.amax(-1, keepdim=True)
+    # s_tmp = torch.exp((qk - m) / math.sqrt(d))
+    # o1 = torch.einsum('bhst,bthd->bshd', s_tmp, v_cache_ref)
+    # lse_ref = torch.logsumexp(qk / math.sqrt(d), -1)
+    # probs = torch.softmax(qk, dim=-1)
+    out_ref, _ = attention_ref(
+        q_ro,
+        k_cache_rep,
+        v_cache_rep,
+        None,
+        key_padding_mask,
+        attn_bias,
+        0.0,
+        None,
+        causal=causal,
+        window_size=window_size,
+        key_leftpad=cache_leftpad,
+    )
+    out_pt, _ = attention_ref(
+        q_ro,
+        k_cache_rep,
+        v_cache_rep,
+        None,
+        key_padding_mask,
+        attn_bias,
+        0.0,
+        None,
+        causal=causal,
+        window_size=window_size,
+        upcast=False,
+        reorder_ops=True,
+        key_leftpad=cache_leftpad,
+    )
+    print(f"Output max diff: {(out - out_ref).abs().max().item()}")
+    print(f"Output mean diff: {(out - out_ref).abs().mean().item()}")
+    print(f"Pytorch max diff: {(out_pt - out_ref).abs().max().item()}")
+    print(f"Pytorch mean diff: {(out_pt - out_ref).abs().mean().item()}")
+
+    # Check that FlashAttention's numerical error is at most twice the numerical error
+    # of a Pytorch implementation.
+    if new_kv:
+        if paged_kv_block_size is None:
+            k_cache_select = (
+                k_cache if not has_batch_idx else k_cache[cache_batch_idx.to(dtype=torch.long)]
+            )
+            v_cache_select = (
+                v_cache if not has_batch_idx else v_cache[cache_batch_idx.to(dtype=torch.long)]
+            )
+        else:
+            k_cache_select = rearrange(
+                k_cache_paged[block_table.to(dtype=torch.long).flatten()],
+                "(b nblocks) block_size ... -> b (nblocks block_size) ...",
+                b=batch_size,
+            )[:, :seqlen_k]
+            v_cache_select = rearrange(
+                v_cache_paged[block_table.to(dtype=torch.long).flatten()],
+                "(b nblocks) block_size ... -> b (nblocks block_size) ...",
+                b=batch_size,
+            )[:, :seqlen_k]
+        assert torch.allclose(k_cache_select, k_cache_ref, rtol=1e-3, atol=1e-3)
+        assert torch.equal(v_cache_select, v_cache_ref)
+    # mult = 3 if f16, bf16 need 4
+    mult = 4 if not alibi else 5
+    assert (out - out_ref).abs().max().item() <= mult * (out_pt - out_ref).abs().max().item() + 1e-5
+
+
+
+@pytest.mark.parametrize("dtype", [torch.float16])
+@pytest.mark.parametrize("causal", [False, True])
+@pytest.mark.parametrize("d", [32, 40, 59, 64, 80, 96, 111, 128, 160, 192, 224, 256])
+@pytest.mark.parametrize(
+    "seqlen_q,seqlen_k",
+    [
+        (1, 239),
+        (239, 1),
+        (3, 799),
+        (799, 3),
+        (1024, 128),
+        (97, 97),
+        (128, 128),
+        (200, 200),
+        (256, 256),
+        (257, 257),
+        (384, 384),
+        (512, 512),
+        (768, 768),
+        # (1024, 1024),
+    ],
+)
+@pytest.mark.parametrize("dropout_p", [0.0, 0.17])
+def test_flash_attn_race_condition(seqlen_q, seqlen_k, d, dropout_p, causal, dtype):
+    device = "cuda"
+    # set seed
+    torch.random.manual_seed(0)
+    batch_size = 60  # Sometimes we need large batch size for the race conditions to trigger
+    nheads = 4
+    q = torch.randn(batch_size, seqlen_q, nheads, d, device=device, dtype=dtype, requires_grad=True)
+    k = torch.randn(batch_size, seqlen_k, nheads, d, device=device, dtype=dtype, requires_grad=True)
+    v = torch.randn(batch_size, seqlen_k, nheads, d, device=device, dtype=dtype, requires_grad=True)
+    torch.random.manual_seed(42)
+    out0, lse0, _ = flash_attn_func(q, k, v, dropout_p, causal=causal, return_attn_probs=True)
+    g = torch.randn_like(out0)
+    if dropout_p == 0 and is_bwd_hdim_supported(d):
+        (
+            dq0,
+            dk0,
+            dv0,
+        ) = torch.autograd.grad(out0, (q, k, v), g)
+        # Numerical error if we just do any arithmetic on dq
+        dq_atol = 2 * ((dq0 + 0.3 - 0.3) - dq0).abs().max().item()
+
+    for i in range(250):
+        torch.random.manual_seed(42)
+        out, lse, _ = flash_attn_func(q, k, v, dropout_p, causal=causal, return_attn_probs=True)
+        assert torch.equal(out, out0)
+        assert torch.equal(lse, lse0)
+
+        if dropout_p == 0:
+            (
+                dq,
+                dk,
+                dv,
+            ) = torch.autograd.grad(out, (q, k, v), g)
+            dq_equal = torch.allclose(dq, dq0, atol=dq_atol)
+            if not dq_equal:
+                print(f"Iter {i}, {dq_atol = }, dQ max diff: {(dq - dq0).abs().max().item()}")
+
+            assert torch.equal(dv, dv0)
+            assert torch.equal(dk, dk0)
+            assert dq_equal
+
+
+@pytest.mark.parametrize("dtype", [torch.float16])
+@pytest.mark.parametrize("causal", [False, True])
+@pytest.mark.parametrize("d", [16, 32, 64])
+@pytest.mark.parametrize("seqlen", [1, 2, 5, 17, 128])
+def test_flash_attn_bwd_overflow(seqlen, d, causal, dtype):
+    """We previously had a bug where not masking elements beyond seqlen_k caused NaN in dQ,
+    in the case where seqlen % 128 != 0.
+    """
+
+    # TODO - 1 or 2 might fail, need to check
+    if seqlen == 1 or seqlen == 2:
+        pytest.skip()
+
+    device = "cuda"
+    # set seed
+    torch.random.manual_seed(0)
+    batch_size = 2
+    nheads = 5
+    q = torch.randn([batch_size, seqlen, nheads, d], dtype=dtype, device="cuda") * 5
+    k, v = [
+        torch.randn([batch_size, seqlen, nheads, d], dtype=dtype, device="cuda") * 3
+        for _ in range(2)
+    ]
+    q.requires_grad_(True)
+    k.requires_grad_(True)
+    v.requires_grad_(True)
+    out = flash_attn_func(q, k, v, causal=causal)
+    g = torch.randn_like(out)
+    out.backward(g)
+    q_pt = q.detach().clone().requires_grad_(True)
+    k_pt = k.detach().clone().requires_grad_(True)
+    v_pt = v.detach().clone().requires_grad_(True)
+    out_pt, _ = attention_ref(q_pt, k_pt, v_pt, causal=causal, upcast=False, reorder_ops=True)
+    out_pt.backward(g)
+    q_ref = q.detach().clone().requires_grad_(True)
+    k_ref = k.detach().clone().requires_grad_(True)
+    v_ref = v.detach().clone().requires_grad_(True)
+    out_ref, attn_ref = attention_ref(q_ref, k_ref, v_ref, causal=causal)
+    out_ref.backward(g)
+    print(f"dQ max diff: {(q.grad - q_ref.grad).abs().max().item()}")
+    print(f"dK max diff: {(k.grad - k_ref.grad).abs().max().item()}")
+    print(f"dV max diff: {(v.grad - v_ref.grad).abs().max().item()}")
+    print(f"dQ Pytorch max diff: {(q_pt.grad - q_ref.grad).abs().max().item()}")
+    print(f"dK Pytorch max diff: {(k_pt.grad - k_ref.grad).abs().max().item()}")
+    print(f"dV Pytorch max diff: {(v_pt.grad - v_ref.grad).abs().max().item()}")
+    assert (out - out_ref).abs().max().item() <= 2 * (out_pt - out_ref).abs().max().item()
+    assert (q.grad - q_ref.grad).abs().max().item() <= 5 * (
+        q_pt.grad - q_ref.grad
+    ).abs().max().item() + 1e-3
+    assert (k.grad - k_ref.grad).abs().max().item() <= 5 * (
+        k_pt.grad - k_ref.grad
+    ).abs().max().item() + 1e-3
+    assert (v.grad - v_ref.grad).abs().max().item() <= 5 * (
+        v_pt.grad - v_ref.grad
+    ).abs().max().item() + 1e-3
+
+
+@pytest.mark.parametrize("dtype", [torch.float16, torch.bfloat16])
+@pytest.mark.parametrize("causal", [False, True])
+@pytest.mark.parametrize("d", [64, 128])
+@pytest.mark.parametrize("seqlen", [97, 128, 200, 256])
+def test_flash_attn_bwd_transpose(seqlen, d, causal, dtype):
+    """We previously had a bug where we were using the wrong strides of dout, which shows up
+    when dout is not contiguous.
+    """
+    device = "cuda"
+    # set seed
+    torch.random.manual_seed(0)
+    batch_size = 5
+    nheads = 2
+    q, k, v = [
+        torch.randn([batch_size, seqlen, nheads, d], dtype=dtype, device="cuda", requires_grad=True)
+        for _ in range(3)
+    ]
+    out = rearrange(flash_attn_func(q, k, v, causal=causal), "b s ... -> s b ...")
+    # So g is not contiguous
+    g = torch.randn(seqlen, 2 * batch_size, nheads, d, dtype=dtype, device="cuda")[:, ::2]
+    out.backward(g)
+    q_pt = q.detach().clone().requires_grad_(True)
+    k_pt = k.detach().clone().requires_grad_(True)
+    v_pt = v.detach().clone().requires_grad_(True)
+    out_pt, attn_pt = attention_ref(q_pt, k_pt, v_pt, causal=causal, upcast=False, reorder_ops=True)
+    out_pt = rearrange(out_pt, "b s ... -> s b ...")
+    out_pt.backward(g)
+    q_ref = q.detach().clone().requires_grad_(True)
+    k_ref = k.detach().clone().requires_grad_(True)
+    v_ref = v.detach().clone().requires_grad_(True)
+    out_ref, attn_ref = attention_ref(q_ref, k_ref, v_ref, causal=causal)
+    out_ref = rearrange(out_ref, "b s ... -> s b ...")
+    out_ref.backward(g)
+    print(f"dQ max diff: {(q.grad - q_ref.grad).abs().max().item()}")
+    print(f"dK max diff: {(k.grad - k_ref.grad).abs().max().item()}")
+    print(f"dV max diff: {(v.grad - v_ref.grad).abs().max().item()}")
+    print(f"dQ Pytorch max diff: {(q_pt.grad - q_ref.grad).abs().max().item()}")
+    print(f"dK Pytorch max diff: {(k_pt.grad - k_ref.grad).abs().max().item()}")
+    print(f"dV Pytorch max diff: {(v_pt.grad - v_ref.grad).abs().max().item()}")
+    assert (out - out_ref).abs().max().item() <= 2 * (out_pt - out_ref).abs().max().item()
+    assert (q.grad - q_ref.grad).abs().max().item() <= 2 * (
+        q_pt.grad - q_ref.grad
+    ).abs().max().item()
+    assert (k.grad - k_ref.grad).abs().max().item() <= 2 * (
+        k_pt.grad - k_ref.grad
+    ).abs().max().item()
+    assert (v.grad - v_ref.grad).abs().max().item() <= 2 * (
+        v_pt.grad - v_ref.grad
+    ).abs().max().item()
+
+
+@pytest.mark.parametrize("dtype", [torch.float16])
+@pytest.mark.parametrize("causal", [False, True])
+@pytest.mark.parametrize("d", [16, 32, 64])
+def test_flash_attn_bwd_varlen_overflow(d, causal, dtype):
+    """We previously had a bug where not masking elements beyond seqlen_k caused NaN in dQ,
+    in the case where seqlen % 128 != 0 or varlen.
+    """
+    device = "cuda"
+    # set seed
+    torch.random.manual_seed(0)
+    nheads = 5
+    q_cuseqlen = torch.tensor([0, 76, 110, 256], device=device, dtype=torch.int32)
+    k_cuseqlen = torch.tensor([0, 1, 2, 3], device=device, dtype=torch.int32)
+    Mq = 256
+    Mk = 3
+
+    q = torch.randn([Mq, nheads, d], dtype=dtype, device=device) * 3
+    k, v = [torch.randn([Mk, nheads, d], dtype=dtype, device=device) * 3 for _ in range(2)]
+    q.requires_grad_(True)
+    k.requires_grad_(True)
+    v.requires_grad_(True)
+
+    out = flash_attn_varlen_func(q, k, v, q_cuseqlen, k_cuseqlen, Mq, Mk, causal=causal)
+    g = torch.randn_like(out)
+    out.backward(g)
+
+    assert not q.grad.isnan().any()
+    assert not k.grad.isnan().any()
+    assert not v.grad.isnan().any()
+
+
+@pytest.mark.parametrize("dtype", [torch.float16, torch.bfloat16])
+@pytest.mark.parametrize("local", [False, True])
+@pytest.mark.parametrize("causal", [False, True])
+@pytest.mark.parametrize("d", [32, 40, 59, 64, 80, 96, 111, 128, 160, 192, 224, 256])
+@pytest.mark.parametrize("swap_sq_sk", [False, True])
+@pytest.mark.parametrize(
+    "seqlen_q,seqlen_k",
+    [
+        (1, 239),
+        (3, 799),
+        (127, 512),
+        (127, 513),
+        (113, 203),
+        (128, 217),
+        (113, 211),
+        (108, 256),
+        (256, 512),
+        (1023, 1024),
+    ],
+)
+def test_flash_attn_deterministic(seqlen_q, seqlen_k, swap_sq_sk, d, causal, local, dtype):
+    if (
+        max(seqlen_q, seqlen_k) >= 2048
+        and torch.cuda.get_device_properties("cuda").total_memory <= 16 * 2**30
+    ):
+        pytest.skip()  # Reference implementation OOM
+    if swap_sq_sk:
+        seqlen_q, seqlen_k = seqlen_k, seqlen_q
+    device = "cuda"
+    # set seed
+    torch.random.manual_seed(0)
+    batch_size = 4
+    nheads = 9
+    window_size = (-1, -1) if not local else torch.randint(0, seqlen_k, (2,))
+    q = torch.randn(batch_size, seqlen_q, nheads, d, device=device, dtype=dtype, requires_grad=True)
+    k = torch.randn(batch_size, seqlen_k, nheads, d, device=device, dtype=dtype, requires_grad=True)
+    v = torch.randn(batch_size, seqlen_k, nheads, d, device=device, dtype=dtype, requires_grad=True)
+    out = flash_attn_func(q, k, v, 0.0, causal=causal, window_size=window_size, deterministic=True)
+
+    g = torch.randn_like(out)
+    dq0, dk0, dv0 = torch.autograd.grad(out, (q, k, v), g, retain_graph=True)
+    for _ in range(50):
+        dq, dk, dv = torch.autograd.grad(out, (q, k, v), g, retain_graph=True)
+        assert torch.equal(dv, dv0)
+        assert torch.equal(dk, dk0)
+        assert torch.equal(dq, dq0)
+
+
+@pytest.mark.parametrize("dtype", [torch.float16, torch.bfloat16])
+@pytest.mark.parametrize("local", [False, True])
+@pytest.mark.parametrize("causal", [False, True])
+@pytest.mark.parametrize("d", [32, 40, 59, 64, 80, 96, 111, 128, 160, 192, 224, 256])
+@pytest.mark.parametrize("swap_sq_sk", [False, True])
+@pytest.mark.parametrize(
+    "seqlen_q,seqlen_k",
+    [
+        (1, 239),
+        (3, 799),
+        (127, 512),
+        (127, 513),
+        (113, 203),
+        (128, 217),
+        (113, 211),
+        (108, 256),
+        (256, 512),
+        (1023, 1024),
+    ],
+)
+def test_flash_attn_varlen_deterministic(seqlen_q, seqlen_k, swap_sq_sk, d, causal, local, dtype):
+    if (
+        max(seqlen_q, seqlen_k) >= 2048
+        and torch.cuda.get_device_properties("cuda").total_memory <= 16 * 2**30
+    ):
+        pytest.skip()  # Reference implementation OOM
+    if swap_sq_sk:
+        seqlen_q, seqlen_k = seqlen_k, seqlen_q
+    device = "cuda"
+    # set seed
+    torch.random.manual_seed(0)
+    batch_size = 2
+    nheads = 9
+    window_size = (-1, -1) if not local else torch.randint(0, seqlen_k, (2,))
+    q = torch.randn(batch_size, seqlen_q, nheads, d, device=device, dtype=dtype, requires_grad=True)
+    k = torch.randn(batch_size, seqlen_k, nheads, d, device=device, dtype=dtype, requires_grad=True)
+    v = torch.randn(batch_size, seqlen_k, nheads, d, device=device, dtype=dtype, requires_grad=True)
+    query_padding_mask = generate_random_padding_mask(seqlen_q, batch_size, device, mode="random")
+    key_padding_mask = generate_random_padding_mask(seqlen_k, batch_size, device, mode="random")
+    (
+        q_unpad,
+        k_unpad,
+        v_unpad,
+        cu_seqlens_q,
+        cu_seqlens_k,
+        max_seqlen_q,
+        max_seqlen_k,
+        q,
+        k,
+        v,
+        output_pad_fn,
+        dq_pad_fn,
+        dk_pad_fn,
+    ) = generate_qkv(q, k, v, query_padding_mask, key_padding_mask, kvpacked=False)
+    out = flash_attn_varlen_func(
+        q_unpad,
+        k_unpad,
+        v_unpad,
+        cu_seqlens_q,
+        cu_seqlens_k,
+        max_seqlen_q,
+        max_seqlen_k,
+        0.0,
+        causal=causal,
+        window_size=window_size,
+        deterministic=True,
+    )
+
+    g = torch.randn_like(out)
+    dq0, dk0, dv0 = torch.autograd.grad(out, (q_unpad, k_unpad, v_unpad), g, retain_graph=True)
+    for _ in range(50):
+        dq, dk, dv = torch.autograd.grad(out, (q_unpad, k_unpad, v_unpad), g, retain_graph=True)
+        assert torch.equal(dv, dv0)
+        assert torch.equal(dk, dk0)
+        assert torch.equal(dq, dq0)
+

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.3_layer_10.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:03:33] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.3_layer_10', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.3_layer_10'], NO_DINO=False, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='4', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.3_layer_10']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:21,  1.56s/it]
 13%|█▎        | 2/15 [00:02<00:12,  1.02it/s]
 20%|██        | 3/15 [00:02<00:07,  1.52it/s]
 27%|██▋       | 4/15 [00:02<00:05,  1.95it/s]
 33%|███▎      | 5/15 [00:02<00:04,  2.33it/s]
 40%|████      | 6/15 [00:03<00:03,  2.61it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.90it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  3.07it/s]
 60%|██████    | 9/15 [00:04<00:01,  3.17it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.25it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.35it/s]
 80%|████████  | 12/15 [00:04<00:00,  3.47it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.54it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.53it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:07,  3.67it/s]
  7%|▋         | 2/30 [00:00<00:07,  3.67it/s]
 10%|█         | 3/30 [00:00<00:07,  3.56it/s]
 13%|█▎        | 4/30 [00:01<00:07,  3.60it/s]
 17%|█▋        | 5/30 [00:01<00:06,  3.73it/s]
 20%|██        | 6/30 [00:01<00:06,  3.59it/s]
 23%|██▎       | 7/30 [00:02<00:07,  3.26it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.25it/s]
 30%|███       | 9/30 [00:02<00:06,  3.37it/s]
 33%|███▎      | 10/30 [00:02<00:05,  3.45it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.59it/s]
 40%|████      | 12/30 [00:03<00:04,  3.61it/s]
 43%|████▎     | 13/30 [00:03<00:04,  3.68it/s]
 47%|████▋     | 14/30 [00:03<00:04,  3.64it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.67it/s]
 53%|█████▎    | 16/30 [00:04<00:03,  3.73it/s]
 57%|█████▋    | 17/30 [00:04<00:03,  3.76it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.55it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.56it/s]
 67%|██████▋   | 20/30 [00:05<00:02,  3.58it/s]
 70%|███████   | 21/30 [00:05<00:02,  3.57it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.56it/s]
 77%|███████▋  | 23/30 [00:06<00:01,  3.67it/s]
 80%|████████  | 24/30 [00:06<00:01,  3.64it/s]
 83%|████████▎ | 25/30 [00:06<00:01,  3.65it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.69it/s]
 90%|█████████ | 27/30 [00:07<00:00,  3.69it/s]
 93%|█████████▎| 28/30 [00:07<00:00,  3.73it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.38it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.59it/s]
 20%|██        | 2/10 [00:00<00:02,  3.55it/s]
 30%|███       | 3/10 [00:00<00:01,  3.70it/s]
 40%|████      | 4/10 [00:01<00:01,  3.65it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.62it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.68it/s]
 70%|███████   | 7/10 [00:01<00:00,  3.67it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.67it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.58it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.50it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.56it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.65it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.68it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.71it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:01,  3.65it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.60it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.63it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.68it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.63it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.68it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.63it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:08,  3.50it/s]
  6%|▋         | 2/32 [00:00<00:08,  3.60it/s]
  9%|▉         | 3/32 [00:00<00:08,  3.57it/s]
 12%|█▎        | 4/32 [00:01<00:07,  3.63it/s]
 16%|█▌        | 5/32 [00:01<00:07,  3.61it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.63it/s]
 22%|██▏       | 7/32 [00:01<00:07,  3.49it/s]
 25%|██▌       | 8/32 [00:02<00:06,  3.53it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.48it/s]
 31%|███▏      | 10/32 [00:02<00:06,  3.48it/s]
 34%|███▍      | 11/32 [00:03<00:06,  3.49it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.53it/s]
 41%|████      | 13/32 [00:03<00:05,  3.58it/s]
 44%|████▍     | 14/32 [00:03<00:04,  3.64it/s]
 47%|████▋     | 15/32 [00:04<00:04,  3.62it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.58it/s]
 53%|█████▎    | 17/32 [00:04<00:04,  3.58it/s]
 56%|█████▋    | 18/32 [00:05<00:03,  3.60it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.59it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.58it/s]
 66%|██████▌   | 21/32 [00:05<00:03,  3.62it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.56it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.55it/s]
 75%|███████▌  | 24/32 [00:06<00:02,  3.63it/s]
 78%|███████▊  | 25/32 [00:06<00:01,  3.69it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.63it/s]
 84%|████████▍ | 27/32 [00:07<00:01,  3.64it/s]
 88%|████████▊ | 28/32 [00:07<00:01,  3.59it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.57it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.58it/s]
 97%|█████████▋| 31/32 [00:08<00:00,  3.65it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:04,  3.63it/s]
 11%|█         | 2/19 [00:00<00:04,  3.58it/s]
 16%|█▌        | 3/19 [00:00<00:04,  3.59it/s]
 21%|██        | 4/19 [00:01<00:04,  3.60it/s]
 26%|██▋       | 5/19 [00:01<00:03,  3.60it/s]
 32%|███▏      | 6/19 [00:01<00:03,  3.54it/s]
 37%|███▋      | 7/19 [00:01<00:03,  3.51it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.58it/s]
 47%|████▋     | 9/19 [00:02<00:02,  3.60it/s]
 53%|█████▎    | 10/19 [00:02<00:02,  3.65it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.68it/s]
 63%|██████▎   | 12/19 [00:03<00:01,  3.63it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.66it/s]
 74%|███████▎  | 14/19 [00:03<00:01,  3.66it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.63it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.62it/s]
 89%|████████▉ | 17/19 [00:04<00:00,  3.60it/s]
 95%|█████████▍| 18/19 [00:04<00:00,  3.62it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.40it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.30it/s]
 11%|█         | 3/27 [00:00<00:07,  3.35it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.51it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.55it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.56it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.60it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.62it/s]
 33%|███▎      | 9/27 [00:02<00:04,  3.62it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.64it/s]
 41%|████      | 11/27 [00:03<00:04,  3.64it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.67it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.63it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.63it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.58it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.58it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.64it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.67it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.67it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.64it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.62it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.61it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.63it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.63it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.61it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.60it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.40it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.63it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.70it/s]
 31%|███       | 4/13 [00:01<00:02,  3.59it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.45it/s]
 46%|████▌     | 6/13 [00:01<00:02,  3.47it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.48it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.59it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.65it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.69it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.70it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.68it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  2.74it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.08it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:01,  3.63it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.62it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.63it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.65it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.62it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.60it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.61it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.58it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.64it/s]
 12%|█▎        | 3/24 [00:00<00:06,  3.35it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.47it/s]
 21%|██        | 5/24 [00:01<00:05,  3.46it/s]
 25%|██▌       | 6/24 [00:01<00:05,  3.44it/s]
 29%|██▉       | 7/24 [00:02<00:04,  3.50it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.51it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.39it/s]
 42%|████▏     | 10/24 [00:02<00:04,  3.49it/s]
 46%|████▌     | 11/24 [00:03<00:03,  3.51it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.52it/s]
 54%|█████▍    | 13/24 [00:03<00:03,  3.55it/s]
 58%|█████▊    | 14/24 [00:04<00:02,  3.52it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.35it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.44it/s]
 71%|███████   | 17/24 [00:04<00:02,  3.47it/s]
 75%|███████▌  | 18/24 [00:05<00:01,  3.49it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.54it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.49it/s]
 88%|████████▊ | 21/24 [00:06<00:00,  3.53it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.56it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.58it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.62it/s]
 22%|██▏       | 2/9 [00:00<00:01,  3.63it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.60it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.69it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.65it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.64it/s]
 78%|███████▊  | 7/9 [00:01<00:00,  3.60it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.60it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.62it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.61it/s]
 11%|█         | 3/27 [00:00<00:06,  3.66it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.66it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.64it/s]
 22%|██▏       | 6/27 [00:01<00:06,  3.41it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.49it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.52it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.57it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.60it/s]
 41%|████      | 11/27 [00:03<00:04,  3.60it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.61it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.64it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.66it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.66it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.65it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.63it/s]
 67%|██████▋   | 18/27 [00:04<00:02,  3.64it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.68it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.63it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.64it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.71it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.68it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.72it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.74it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.73it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  3.64it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.76it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.82it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.79it/s]
 42%|████▏     | 5/12 [00:01<00:01,  3.68it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.71it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.64it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.65it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.61it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.61it/s]
 92%|█████████▏| 11/12 [00:03<00:00,  3.58it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.40it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.46it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.50it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.48it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.48it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.50it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.37it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.58it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.58it/s]
 31%|███       | 4/13 [00:01<00:02,  3.55it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.51it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.54it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.56it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.54it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.55it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.60it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.55it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.61it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.49it/s]
 10%|█         | 2/20 [00:00<00:05,  3.53it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.59it/s]
 20%|██        | 4/20 [00:01<00:04,  3.60it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.58it/s]
 30%|███       | 6/20 [00:01<00:03,  3.59it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.63it/s]
 40%|████      | 8/20 [00:02<00:03,  3.59it/s]
 45%|████▌     | 9/20 [00:02<00:03,  3.57it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.60it/s]
 55%|█████▌    | 11/20 [00:03<00:02,  3.60it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.61it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.58it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.58it/s]
 75%|███████▌  | 15/20 [00:04<00:01,  3.60it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.58it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.66it/s]
 90%|█████████ | 18/20 [00:05<00:00,  3.61it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.60it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:04,  3.49it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.50it/s]
 20%|██        | 3/15 [00:00<00:03,  3.55it/s]
 27%|██▋       | 4/15 [00:01<00:03,  3.50it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.54it/s]
 40%|████      | 6/15 [00:01<00:02,  3.54it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.63it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.66it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.67it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.68it/s]
 73%|███████▎  | 11/15 [00:03<00:01,  3.67it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.65it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.62it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.60it/s]
+
+[2025-07-11 17:04:59] Weighted Per image PCK0.10: 75.93%, image PCK0.05: 58.41%, image PCK0.01: 82.86%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.3_layer_10_no_dino.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:03:33] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.3_layer_10_no_dino', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.3_layer_10'], NO_DINO=True, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='5', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.3_layer_10']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:21,  1.51s/it]
 13%|█▎        | 2/15 [00:01<00:10,  1.23it/s]
 20%|██        | 3/15 [00:02<00:07,  1.54it/s]
 27%|██▋       | 4/15 [00:02<00:05,  2.00it/s]
 33%|███▎      | 5/15 [00:02<00:04,  2.36it/s]
 40%|████      | 6/15 [00:03<00:03,  2.68it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.80it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  2.87it/s]
 60%|██████    | 9/15 [00:04<00:01,  3.06it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.15it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.19it/s]
 80%|████████  | 12/15 [00:04<00:00,  3.23it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.28it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.29it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:10,  2.77it/s]
  7%|▋         | 2/30 [00:00<00:10,  2.76it/s]
 10%|█         | 3/30 [00:01<00:08,  3.03it/s]
 13%|█▎        | 4/30 [00:01<00:08,  2.98it/s]
 17%|█▋        | 5/30 [00:01<00:08,  3.10it/s]
 20%|██        | 6/30 [00:01<00:07,  3.11it/s]
 23%|██▎       | 7/30 [00:02<00:07,  3.15it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.28it/s]
 30%|███       | 9/30 [00:02<00:06,  3.31it/s]
 33%|███▎      | 10/30 [00:03<00:05,  3.42it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.41it/s]
 40%|████      | 12/30 [00:03<00:05,  3.41it/s]
 43%|████▎     | 13/30 [00:04<00:04,  3.41it/s]
 47%|████▋     | 14/30 [00:04<00:04,  3.51it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.54it/s]
 53%|█████▎    | 16/30 [00:04<00:03,  3.54it/s]
 57%|█████▋    | 17/30 [00:05<00:03,  3.49it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.50it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.50it/s]
 67%|██████▋   | 20/30 [00:05<00:02,  3.50it/s]
 70%|███████   | 21/30 [00:06<00:02,  3.52it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.59it/s]
 77%|███████▋  | 23/30 [00:06<00:02,  3.46it/s]
 80%|████████  | 24/30 [00:07<00:01,  3.44it/s]
 83%|████████▎ | 25/30 [00:07<00:01,  3.43it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.46it/s]
 90%|█████████ | 27/30 [00:07<00:00,  3.50it/s]
 93%|█████████▎| 28/30 [00:08<00:00,  3.47it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.41it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:03,  2.78it/s]
 20%|██        | 2/10 [00:00<00:02,  3.15it/s]
 30%|███       | 3/10 [00:00<00:02,  3.29it/s]
 40%|████      | 4/10 [00:01<00:01,  3.36it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.40it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.41it/s]
 70%|███████   | 7/10 [00:02<00:00,  3.49it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.38it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.42it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.72it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.64it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.41it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.30it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.40it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.35it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.61it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.61it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.62it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.61it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.50it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.56it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:10,  3.05it/s]
  6%|▋         | 2/32 [00:00<00:09,  3.31it/s]
  9%|▉         | 3/32 [00:00<00:08,  3.34it/s]
 12%|█▎        | 4/32 [00:01<00:08,  3.40it/s]
 16%|█▌        | 5/32 [00:01<00:07,  3.46it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.41it/s]
 22%|██▏       | 7/32 [00:02<00:07,  3.45it/s]
 25%|██▌       | 8/32 [00:02<00:07,  3.38it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.30it/s]
 31%|███▏      | 10/32 [00:02<00:06,  3.43it/s]
 34%|███▍      | 11/32 [00:03<00:06,  3.46it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.37it/s]
 41%|████      | 13/32 [00:03<00:05,  3.41it/s]
 44%|████▍     | 14/32 [00:04<00:05,  3.43it/s]
 47%|████▋     | 15/32 [00:04<00:04,  3.43it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.50it/s]
 53%|█████▎    | 17/32 [00:04<00:04,  3.46it/s]
 56%|█████▋    | 18/32 [00:05<00:04,  3.37it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.46it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.44it/s]
 66%|██████▌   | 21/32 [00:06<00:03,  3.44it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.55it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.61it/s]
 75%|███████▌  | 24/32 [00:06<00:02,  3.57it/s]
 78%|███████▊  | 25/32 [00:07<00:02,  3.49it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.55it/s]
 84%|████████▍ | 27/32 [00:07<00:01,  3.42it/s]
 88%|████████▊ | 28/32 [00:08<00:01,  3.52it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.53it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.58it/s]
 97%|█████████▋| 31/32 [00:08<00:00,  3.47it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:06,  2.76it/s]
 11%|█         | 2/19 [00:00<00:05,  3.20it/s]
 16%|█▌        | 3/19 [00:00<00:05,  3.07it/s]
 21%|██        | 4/19 [00:01<00:04,  3.18it/s]
 26%|██▋       | 5/19 [00:01<00:04,  3.11it/s]
 32%|███▏      | 6/19 [00:01<00:04,  3.15it/s]
 37%|███▋      | 7/19 [00:02<00:03,  3.30it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.20it/s]
 47%|████▋     | 9/19 [00:02<00:03,  3.25it/s]
 53%|█████▎    | 10/19 [00:03<00:02,  3.28it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.36it/s]
 63%|██████▎   | 12/19 [00:03<00:02,  3.44it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.46it/s]
 74%|███████▎  | 14/19 [00:04<00:01,  3.41it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.41it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.35it/s]
 89%|████████▉ | 17/19 [00:05<00:00,  3.37it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.42it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.46it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.46it/s]
 11%|█         | 3/27 [00:00<00:06,  3.60it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.58it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.57it/s]
 22%|██▏       | 6/27 [00:01<00:06,  3.35it/s]
 26%|██▌       | 7/27 [00:02<00:05,  3.42it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.44it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.56it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.59it/s]
 41%|████      | 11/27 [00:03<00:04,  3.45it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.42it/s]
 48%|████▊     | 13/27 [00:03<00:04,  3.25it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.36it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.37it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.44it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.39it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.43it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.38it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.51it/s]
 78%|███████▊  | 21/27 [00:06<00:01,  3.58it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.66it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.59it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.61it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.51it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.30it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:04,  2.76it/s]
 15%|█▌        | 2/13 [00:00<00:03,  2.84it/s]
 23%|██▎       | 3/13 [00:01<00:03,  3.00it/s]
 31%|███       | 4/13 [00:01<00:02,  3.17it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.08it/s]
 46%|████▌     | 6/13 [00:01<00:02,  3.21it/s]
 54%|█████▍    | 7/13 [00:02<00:01,  3.28it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.36it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.37it/s]
 77%|███████▋  | 10/13 [00:03<00:00,  3.38it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.44it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.50it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.46it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.52it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  2.86it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.22it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.41it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.45it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.47it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.42it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.49it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.44it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.53it/s]
 12%|█▎        | 3/24 [00:00<00:05,  3.62it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.56it/s]
 21%|██        | 5/24 [00:01<00:05,  3.56it/s]
 25%|██▌       | 6/24 [00:01<00:05,  3.57it/s]
 29%|██▉       | 7/24 [00:01<00:04,  3.50it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.51it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.54it/s]
 42%|████▏     | 10/24 [00:02<00:04,  3.43it/s]
 46%|████▌     | 11/24 [00:03<00:03,  3.48it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.53it/s]
 54%|█████▍    | 13/24 [00:03<00:03,  3.57it/s]
 58%|█████▊    | 14/24 [00:03<00:02,  3.60it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.60it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.62it/s]
 71%|███████   | 17/24 [00:04<00:02,  3.46it/s]
 75%|███████▌  | 18/24 [00:05<00:01,  3.37it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.38it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.50it/s]
 88%|████████▊ | 21/24 [00:05<00:00,  3.54it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.54it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.56it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:03,  2.60it/s]
 22%|██▏       | 2/9 [00:00<00:02,  2.77it/s]
 33%|███▎      | 3/9 [00:01<00:01,  3.04it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.17it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.32it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.41it/s]
 78%|███████▊  | 7/9 [00:02<00:00,  3.51it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.56it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:10,  2.45it/s]
  7%|▋         | 2/27 [00:00<00:09,  2.64it/s]
 11%|█         | 3/27 [00:01<00:08,  2.96it/s]
 15%|█▍        | 4/27 [00:01<00:07,  3.17it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.31it/s]
 22%|██▏       | 6/27 [00:01<00:06,  3.47it/s]
 26%|██▌       | 7/27 [00:02<00:05,  3.49it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.45it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.49it/s]
 37%|███▋      | 10/27 [00:03<00:04,  3.43it/s]
 41%|████      | 11/27 [00:03<00:04,  3.44it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.50it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.54it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.56it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.46it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.41it/s]
 63%|██████▎   | 17/27 [00:05<00:02,  3.49it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.50it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.44it/s]
 74%|███████▍  | 20/27 [00:05<00:02,  3.42it/s]
 78%|███████▊  | 21/27 [00:06<00:01,  3.48it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.48it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.50it/s]
 89%|████████▉ | 24/27 [00:07<00:00,  3.42it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.42it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.43it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:04,  2.37it/s]
 17%|█▋        | 2/12 [00:00<00:03,  2.90it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.15it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.28it/s]
 42%|████▏     | 5/12 [00:01<00:02,  3.37it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.48it/s]
 58%|█████▊    | 7/12 [00:02<00:01,  3.57it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.58it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.63it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.65it/s]
 92%|█████████▏| 11/12 [00:03<00:00,  3.68it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.83it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.86it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.82it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.84it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.89it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.81it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.27it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.50it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.56it/s]
 31%|███       | 4/13 [00:01<00:02,  3.57it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.61it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.61it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.60it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.62it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.65it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.64it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.71it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.65it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.73it/s]
 10%|█         | 2/20 [00:00<00:04,  3.78it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.77it/s]
 20%|██        | 4/20 [00:01<00:04,  3.78it/s]
 25%|██▌       | 5/20 [00:01<00:03,  3.84it/s]
 30%|███       | 6/20 [00:01<00:03,  3.81it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.75it/s]
 40%|████      | 8/20 [00:02<00:03,  3.83it/s]
 45%|████▌     | 9/20 [00:02<00:02,  3.83it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.78it/s]
 55%|█████▌    | 11/20 [00:02<00:02,  3.84it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.80it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.87it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.87it/s]
 75%|███████▌  | 15/20 [00:03<00:01,  3.77it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.84it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.77it/s]
 90%|█████████ | 18/20 [00:04<00:00,  3.77it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.74it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:04,  3.16it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.50it/s]
 20%|██        | 3/15 [00:00<00:03,  3.61it/s]
 27%|██▋       | 4/15 [00:01<00:02,  3.72it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.72it/s]
 40%|████      | 6/15 [00:01<00:02,  3.82it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.75it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.72it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.72it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.71it/s]
 73%|███████▎  | 11/15 [00:02<00:01,  3.64it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.59it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.61it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.68it/s]
+
+[2025-07-11 17:05:02] Weighted Per image PCK0.10: 60.83%, image PCK0.05: 41.10%, image PCK0.01: 74.43%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.3_layer_13.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:05:07] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.3_layer_13', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.3_layer_13'], NO_DINO=False, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='6', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.3_layer_13']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:20,  1.44s/it]
 13%|█▎        | 2/15 [00:01<00:11,  1.11it/s]
 20%|██        | 3/15 [00:02<00:08,  1.48it/s]
 27%|██▋       | 4/15 [00:02<00:05,  1.92it/s]
 33%|███▎      | 5/15 [00:02<00:04,  2.24it/s]
 40%|████      | 6/15 [00:03<00:03,  2.55it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.73it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  2.88it/s]
 60%|██████    | 9/15 [00:04<00:01,  3.01it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.13it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.22it/s]
 80%|████████  | 12/15 [00:05<00:00,  3.23it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.34it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.39it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:08,  3.26it/s]
  7%|▋         | 2/30 [00:00<00:08,  3.32it/s]
 10%|█         | 3/30 [00:00<00:07,  3.48it/s]
 13%|█▎        | 4/30 [00:01<00:07,  3.42it/s]
 17%|█▋        | 5/30 [00:01<00:07,  3.41it/s]
 20%|██        | 6/30 [00:01<00:07,  3.39it/s]
 23%|██▎       | 7/30 [00:02<00:06,  3.43it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.51it/s]
 30%|███       | 9/30 [00:02<00:06,  3.50it/s]
 33%|███▎      | 10/30 [00:02<00:05,  3.56it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.53it/s]
 40%|████      | 12/30 [00:03<00:05,  3.54it/s]
 43%|████▎     | 13/30 [00:03<00:04,  3.53it/s]
 47%|████▋     | 14/30 [00:03<00:04,  3.62it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.62it/s]
 53%|█████▎    | 16/30 [00:04<00:03,  3.60it/s]
 57%|█████▋    | 17/30 [00:04<00:03,  3.57it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.58it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.60it/s]
 67%|██████▋   | 20/30 [00:05<00:02,  3.59it/s]
 70%|███████   | 21/30 [00:05<00:02,  3.62it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.66it/s]
 77%|███████▋  | 23/30 [00:06<00:01,  3.61it/s]
 80%|████████  | 24/30 [00:06<00:01,  3.65it/s]
 83%|████████▎ | 25/30 [00:07<00:01,  3.61it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.60it/s]
 90%|█████████ | 27/30 [00:07<00:00,  3.60it/s]
 93%|█████████▎| 28/30 [00:07<00:00,  3.57it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.54it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.63it/s]
 20%|██        | 2/10 [00:00<00:02,  3.47it/s]
 30%|███       | 3/10 [00:00<00:02,  3.31it/s]
 40%|████      | 4/10 [00:01<00:01,  3.31it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.36it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.36it/s]
 70%|███████   | 7/10 [00:02<00:00,  3.42it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.45it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.46it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.81it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.68it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.70it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.65it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.60it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:01,  3.51it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.77it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.59it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.50it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.48it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.51it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.58it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:08,  3.76it/s]
  6%|▋         | 2/32 [00:00<00:08,  3.63it/s]
  9%|▉         | 3/32 [00:00<00:08,  3.55it/s]
 12%|█▎        | 4/32 [00:01<00:08,  3.33it/s]
 16%|█▌        | 5/32 [00:01<00:07,  3.39it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.42it/s]
 22%|██▏       | 7/32 [00:02<00:07,  3.49it/s]
 25%|██▌       | 8/32 [00:02<00:06,  3.49it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.51it/s]
 31%|███▏      | 10/32 [00:02<00:06,  3.60it/s]
 34%|███▍      | 11/32 [00:03<00:05,  3.64it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.64it/s]
 41%|████      | 13/32 [00:03<00:05,  3.60it/s]
 44%|████▍     | 14/32 [00:03<00:05,  3.57it/s]
 47%|████▋     | 15/32 [00:04<00:04,  3.63it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.67it/s]
 53%|█████▎    | 17/32 [00:04<00:03,  3.76it/s]
 56%|█████▋    | 18/32 [00:05<00:03,  3.63it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.66it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.61it/s]
 66%|██████▌   | 21/32 [00:05<00:03,  3.59it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.69it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.72it/s]
 75%|███████▌  | 24/32 [00:06<00:02,  3.62it/s]
 78%|███████▊  | 25/32 [00:06<00:01,  3.53it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.59it/s]
 84%|████████▍ | 27/32 [00:07<00:01,  3.60it/s]
 88%|████████▊ | 28/32 [00:07<00:01,  3.67it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.71it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.74it/s]
 97%|█████████▋| 31/32 [00:08<00:00,  3.68it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:05,  3.53it/s]
 11%|█         | 2/19 [00:00<00:04,  3.69it/s]
 16%|█▌        | 3/19 [00:00<00:04,  3.65it/s]
 21%|██        | 4/19 [00:01<00:04,  3.61it/s]
 26%|██▋       | 5/19 [00:01<00:03,  3.58it/s]
 32%|███▏      | 6/19 [00:01<00:03,  3.62it/s]
 37%|███▋      | 7/19 [00:01<00:03,  3.55it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.51it/s]
 47%|████▋     | 9/19 [00:02<00:02,  3.49it/s]
 53%|█████▎    | 10/19 [00:02<00:02,  3.46it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.48it/s]
 63%|██████▎   | 12/19 [00:03<00:01,  3.54it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.53it/s]
 74%|███████▎  | 14/19 [00:03<00:01,  3.56it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.62it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.68it/s]
 89%|████████▉ | 17/19 [00:04<00:00,  3.67it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.63it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.47it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.57it/s]
 11%|█         | 3/27 [00:00<00:06,  3.64it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.56it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.58it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.58it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.58it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.52it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.59it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.58it/s]
 41%|████      | 11/27 [00:03<00:04,  3.58it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.54it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.61it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.62it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.67it/s]
 59%|█████▉    | 16/27 [00:04<00:02,  3.73it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.68it/s]
 67%|██████▋   | 18/27 [00:04<00:02,  3.63it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.64it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.72it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.74it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.76it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.72it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.68it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.67it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.66it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.28it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.38it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.46it/s]
 31%|███       | 4/13 [00:01<00:02,  3.51it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.57it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.63it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.63it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.59it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.57it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.57it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.56it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.62it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.45it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.54it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.37it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.42it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.51it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.51it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.57it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.59it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.62it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.49it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.55it/s]
 12%|█▎        | 3/24 [00:00<00:05,  3.63it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.56it/s]
 21%|██        | 5/24 [00:01<00:05,  3.58it/s]
 25%|██▌       | 6/24 [00:01<00:04,  3.60it/s]
 29%|██▉       | 7/24 [00:01<00:04,  3.61it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.59it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.58it/s]
 42%|████▏     | 10/24 [00:02<00:03,  3.54it/s]
 46%|████▌     | 11/24 [00:03<00:03,  3.55it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.56it/s]
 54%|█████▍    | 13/24 [00:03<00:03,  3.54it/s]
 58%|█████▊    | 14/24 [00:03<00:02,  3.54it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.59it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.60it/s]
 71%|███████   | 17/24 [00:04<00:01,  3.61it/s]
 75%|███████▌  | 18/24 [00:05<00:01,  3.62it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.62it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.68it/s]
 88%|████████▊ | 21/24 [00:05<00:00,  3.67it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.66it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.64it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.68it/s]
 22%|██▏       | 2/9 [00:00<00:01,  3.70it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.64it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.61it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.67it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.71it/s]
 78%|███████▊  | 7/9 [00:01<00:00,  3.79it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.78it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.62it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.64it/s]
 11%|█         | 3/27 [00:00<00:06,  3.65it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.68it/s]
 19%|█▊        | 5/27 [00:01<00:05,  3.68it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.79it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.72it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.72it/s]
 33%|███▎      | 9/27 [00:02<00:04,  3.75it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.74it/s]
 41%|████      | 11/27 [00:02<00:04,  3.74it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.71it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.70it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.71it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.70it/s]
 59%|█████▉    | 16/27 [00:04<00:02,  3.69it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.71it/s]
 67%|██████▋   | 18/27 [00:04<00:02,  3.71it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.67it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.71it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.70it/s]
 81%|████████▏ | 22/27 [00:05<00:01,  3.66it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.67it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.67it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.62it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.63it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  3.64it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.57it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.56it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.54it/s]
 42%|████▏     | 5/12 [00:01<00:01,  3.64it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.66it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.71it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.68it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.71it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.70it/s]
 92%|█████████▏| 11/12 [00:02<00:00,  3.72it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.80it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.80it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.72it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.73it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.77it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.67it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.68it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.63it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.63it/s]
 31%|███       | 4/13 [00:01<00:02,  3.62it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.64it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.66it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.61it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.61it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.63it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.60it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.66it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.60it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.59it/s]
 10%|█         | 2/20 [00:00<00:04,  3.63it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.60it/s]
 20%|██        | 4/20 [00:01<00:04,  3.60it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.69it/s]
 30%|███       | 6/20 [00:01<00:03,  3.67it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.64it/s]
 40%|████      | 8/20 [00:02<00:03,  3.71it/s]
 45%|████▌     | 9/20 [00:02<00:02,  3.75it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.71it/s]
 55%|█████▌    | 11/20 [00:02<00:02,  3.77it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.73it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.78it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.78it/s]
 75%|███████▌  | 15/20 [00:04<00:01,  3.72it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.78it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.70it/s]
 90%|█████████ | 18/20 [00:04<00:00,  3.72it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.68it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:03,  3.65it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.80it/s]
 20%|██        | 3/15 [00:00<00:03,  3.75it/s]
 27%|██▋       | 4/15 [00:01<00:02,  3.76it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.72it/s]
 40%|████      | 6/15 [00:01<00:02,  3.80it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.70it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.66it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.64it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.60it/s]
 73%|███████▎  | 11/15 [00:02<00:01,  3.62it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.64it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.69it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.72it/s]
+
+[2025-07-11 17:06:32] Weighted Per image PCK0.10: 69.53%, image PCK0.05: 50.55%, image PCK0.01: 78.63%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.3_layer_15.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:06:36] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.3_layer_15', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.3_layer_15'], NO_DINO=False, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='4', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.3_layer_15']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:21,  1.56s/it]
 13%|█▎        | 2/15 [00:02<00:12,  1.04it/s]
 20%|██        | 3/15 [00:02<00:07,  1.55it/s]
 27%|██▋       | 4/15 [00:02<00:05,  1.96it/s]
 33%|███▎      | 5/15 [00:03<00:04,  2.19it/s]
 40%|████      | 6/15 [00:03<00:03,  2.42it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.71it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  2.89it/s]
 60%|██████    | 9/15 [00:04<00:01,  3.06it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.22it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.38it/s]
 80%|████████  | 12/15 [00:05<00:00,  3.53it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.61it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.52it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:07,  3.68it/s]
  7%|▋         | 2/30 [00:00<00:08,  3.28it/s]
 10%|█         | 3/30 [00:00<00:08,  3.29it/s]
 13%|█▎        | 4/30 [00:01<00:07,  3.25it/s]
 17%|█▋        | 5/30 [00:01<00:07,  3.36it/s]
 20%|██        | 6/30 [00:01<00:07,  3.39it/s]
 23%|██▎       | 7/30 [00:02<00:06,  3.41it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.39it/s]
 30%|███       | 9/30 [00:02<00:06,  3.45it/s]
 33%|███▎      | 10/30 [00:02<00:05,  3.45it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.52it/s]
 40%|████      | 12/30 [00:03<00:05,  3.48it/s]
 43%|████▎     | 13/30 [00:03<00:04,  3.53it/s]
 47%|████▋     | 14/30 [00:04<00:04,  3.49it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.52it/s]
 53%|█████▎    | 16/30 [00:04<00:03,  3.53it/s]
 57%|█████▋    | 17/30 [00:04<00:03,  3.57it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.52it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.51it/s]
 67%|██████▋   | 20/30 [00:05<00:02,  3.46it/s]
 70%|███████   | 21/30 [00:06<00:02,  3.38it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.35it/s]
 77%|███████▋  | 23/30 [00:06<00:02,  3.40it/s]
 80%|████████  | 24/30 [00:06<00:01,  3.35it/s]
 83%|████████▎ | 25/30 [00:07<00:01,  3.38it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.44it/s]
 90%|█████████ | 27/30 [00:07<00:00,  3.41it/s]
 93%|█████████▎| 28/30 [00:08<00:00,  3.46it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.49it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.58it/s]
 20%|██        | 2/10 [00:00<00:02,  3.49it/s]
 30%|███       | 3/10 [00:00<00:02,  3.50it/s]
 40%|████      | 4/10 [00:01<00:01,  3.21it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.29it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.42it/s]
 70%|███████   | 7/10 [00:02<00:00,  3.46it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.53it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.50it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.30it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.39it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.51it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.54it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.64it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:01,  3.63it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.52it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.53it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.52it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.54it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.56it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.52it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:09,  3.34it/s]
  6%|▋         | 2/32 [00:00<00:08,  3.49it/s]
  9%|▉         | 3/32 [00:00<00:08,  3.49it/s]
 12%|█▎        | 4/32 [00:01<00:07,  3.55it/s]
 16%|█▌        | 5/32 [00:01<00:07,  3.54it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.61it/s]
 22%|██▏       | 7/32 [00:01<00:06,  3.61it/s]
 25%|██▌       | 8/32 [00:02<00:06,  3.68it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.67it/s]
 31%|███▏      | 10/32 [00:02<00:06,  3.59it/s]
 34%|███▍      | 11/32 [00:03<00:05,  3.56it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.54it/s]
 41%|████      | 13/32 [00:03<00:05,  3.60it/s]
 44%|████▍     | 14/32 [00:03<00:04,  3.64it/s]
 47%|████▋     | 15/32 [00:04<00:04,  3.63it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.57it/s]
 53%|█████▎    | 17/32 [00:04<00:04,  3.48it/s]
 56%|█████▋    | 18/32 [00:05<00:04,  3.41it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.43it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.50it/s]
 66%|██████▌   | 21/32 [00:05<00:03,  3.56it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.50it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.49it/s]
 75%|███████▌  | 24/32 [00:06<00:02,  3.56it/s]
 78%|███████▊  | 25/32 [00:07<00:01,  3.61it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.56it/s]
 84%|████████▍ | 27/32 [00:07<00:01,  3.58it/s]
 88%|████████▊ | 28/32 [00:07<00:01,  3.55it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.53it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.52it/s]
 97%|█████████▋| 31/32 [00:08<00:00,  3.57it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:05,  3.57it/s]
 11%|█         | 2/19 [00:00<00:04,  3.54it/s]
 16%|█▌        | 3/19 [00:00<00:04,  3.58it/s]
 21%|██        | 4/19 [00:01<00:04,  3.59it/s]
 26%|██▋       | 5/19 [00:01<00:03,  3.57it/s]
 32%|███▏      | 6/19 [00:01<00:03,  3.49it/s]
 37%|███▋      | 7/19 [00:01<00:03,  3.46it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.48it/s]
 47%|████▋     | 9/19 [00:02<00:02,  3.46it/s]
 53%|█████▎    | 10/19 [00:02<00:02,  3.52it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.55it/s]
 63%|██████▎   | 12/19 [00:03<00:02,  3.48it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.51it/s]
 74%|███████▎  | 14/19 [00:03<00:01,  3.52it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.51it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.53it/s]
 89%|████████▉ | 17/19 [00:04<00:00,  3.52it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.56it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.54it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.49it/s]
 11%|█         | 3/27 [00:00<00:06,  3.49it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.59it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.54it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.53it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.49it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.51it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.47it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.52it/s]
 41%|████      | 11/27 [00:03<00:04,  3.56it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.58it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.54it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.51it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.49it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.50it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.55it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.60it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.59it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.55it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.51it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.51it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.52it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.53it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.55it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.55it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.41it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.59it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.66it/s]
 31%|███       | 4/13 [00:01<00:02,  3.64it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.56it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.54it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.56it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.63it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.66it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.68it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.71it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.58it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.64it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.56it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:01,  3.56it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.61it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.63it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.70it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.65it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.62it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.63it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.53it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.60it/s]
 12%|█▎        | 3/24 [00:00<00:05,  3.58it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.65it/s]
 21%|██        | 5/24 [00:01<00:05,  3.59it/s]
 25%|██▌       | 6/24 [00:01<00:05,  3.56it/s]
 29%|██▉       | 7/24 [00:01<00:04,  3.52it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.52it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.53it/s]
 42%|████▏     | 10/24 [00:02<00:03,  3.59it/s]
 46%|████▌     | 11/24 [00:03<00:03,  3.62it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.62it/s]
 54%|█████▍    | 13/24 [00:03<00:03,  3.60it/s]
 58%|█████▊    | 14/24 [00:03<00:02,  3.56it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.57it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.60it/s]
 71%|███████   | 17/24 [00:04<00:01,  3.61it/s]
 75%|███████▌  | 18/24 [00:05<00:01,  3.60it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.62it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.58it/s]
 88%|████████▊ | 21/24 [00:05<00:00,  3.58it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.55it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.53it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.51it/s]
 22%|██▏       | 2/9 [00:00<00:01,  3.57it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.55it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.65it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.59it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.57it/s]
 78%|███████▊  | 7/9 [00:01<00:00,  3.54it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.56it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:06,  3.77it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.70it/s]
 11%|█         | 3/27 [00:00<00:06,  3.52it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.51it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.55it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.52it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.56it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.56it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.57it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.50it/s]
 41%|████      | 11/27 [00:03<00:04,  3.52it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.54it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.59it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.59it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.57it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.57it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.57it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.58it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.60it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.59it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.60it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.67it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.65it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.66it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.60it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.56it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  3.62it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.73it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.69it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.71it/s]
 42%|████▏     | 5/12 [00:01<00:01,  3.61it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.62it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.58it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.59it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.54it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.55it/s]
 92%|█████████▏| 11/12 [00:03<00:00,  3.51it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.28it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.35it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.43it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.43it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.39it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.39it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.40it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.52it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.54it/s]
 31%|███       | 4/13 [00:01<00:02,  3.53it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.51it/s]
 46%|████▌     | 6/13 [00:01<00:02,  3.49it/s]
 54%|█████▍    | 7/13 [00:02<00:01,  3.46it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.43it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.42it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.48it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.42it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.48it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.26it/s]
 10%|█         | 2/20 [00:00<00:05,  3.41it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.48it/s]
 20%|██        | 4/20 [00:01<00:04,  3.50it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.44it/s]
 30%|███       | 6/20 [00:01<00:04,  3.50it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.58it/s]
 40%|████      | 8/20 [00:02<00:03,  3.53it/s]
 45%|████▌     | 9/20 [00:02<00:03,  3.50it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.54it/s]
 55%|█████▌    | 11/20 [00:03<00:02,  3.54it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.55it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.54it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.53it/s]
 75%|███████▌  | 15/20 [00:04<00:01,  3.58it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.54it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.62it/s]
 90%|█████████ | 18/20 [00:05<00:00,  3.58it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.60it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:04,  2.91it/s]
 13%|█▎        | 2/15 [00:00<00:04,  3.22it/s]
 20%|██        | 3/15 [00:00<00:03,  3.37it/s]
 27%|██▋       | 4/15 [00:01<00:03,  3.40it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.44it/s]
 40%|████      | 6/15 [00:01<00:02,  3.43it/s]
 47%|████▋     | 7/15 [00:02<00:02,  3.52it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.56it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.59it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.62it/s]
 73%|███████▎  | 11/15 [00:03<00:01,  3.62it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.62it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.59it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.57it/s]
+
+[2025-07-11 17:08:03] Weighted Per image PCK0.10: 65.37%, image PCK0.05: 47.27%, image PCK0.01: 73.21%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.3_layer_15_no_dino.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:06:37] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.3_layer_15_no_dino', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.3_layer_15'], NO_DINO=True, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='5', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.3_layer_15']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:19,  1.38s/it]
 13%|█▎        | 2/15 [00:01<00:11,  1.13it/s]
 20%|██        | 3/15 [00:02<00:07,  1.52it/s]
 27%|██▋       | 4/15 [00:02<00:06,  1.83it/s]
 33%|███▎      | 5/15 [00:02<00:04,  2.20it/s]
 40%|████      | 6/15 [00:03<00:03,  2.56it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.80it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  2.98it/s]
 60%|██████    | 9/15 [00:04<00:01,  3.15it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.27it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.33it/s]
 80%|████████  | 12/15 [00:04<00:00,  3.38it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.40it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.46it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:08,  3.23it/s]
  7%|▋         | 2/30 [00:00<00:08,  3.35it/s]
 10%|█         | 3/30 [00:00<00:07,  3.51it/s]
 13%|█▎        | 4/30 [00:01<00:07,  3.49it/s]
 17%|█▋        | 5/30 [00:01<00:07,  3.45it/s]
 20%|██        | 6/30 [00:01<00:06,  3.46it/s]
 23%|██▎       | 7/30 [00:02<00:06,  3.47it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.55it/s]
 30%|███       | 9/30 [00:02<00:05,  3.54it/s]
 33%|███▎      | 10/30 [00:02<00:05,  3.56it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.52it/s]
 40%|████      | 12/30 [00:03<00:05,  3.50it/s]
 43%|████▎     | 13/30 [00:03<00:04,  3.49it/s]
 47%|████▋     | 14/30 [00:03<00:04,  3.57it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.56it/s]
 53%|█████▎    | 16/30 [00:04<00:03,  3.53it/s]
 57%|█████▋    | 17/30 [00:04<00:03,  3.50it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.51it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.52it/s]
 67%|██████▋   | 20/30 [00:05<00:02,  3.53it/s]
 70%|███████   | 21/30 [00:05<00:02,  3.57it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.61it/s]
 77%|███████▋  | 23/30 [00:06<00:01,  3.56it/s]
 80%|████████  | 24/30 [00:06<00:01,  3.58it/s]
 83%|████████▎ | 25/30 [00:07<00:01,  3.53it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.44it/s]
 90%|█████████ | 27/30 [00:07<00:00,  3.47it/s]
 93%|█████████▎| 28/30 [00:07<00:00,  3.47it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.47it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.29it/s]
 20%|██        | 2/10 [00:00<00:02,  3.43it/s]
 30%|███       | 3/10 [00:00<00:02,  3.41it/s]
 40%|████      | 4/10 [00:01<00:01,  3.47it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.51it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.48it/s]
 70%|███████   | 7/10 [00:02<00:00,  3.50it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.50it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.52it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.57it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.57it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.54it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.50it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.52it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.45it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.71it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.64it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.62it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.62it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.59it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.69it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:08,  3.75it/s]
  6%|▋         | 2/32 [00:00<00:10,  2.75it/s]
  9%|▉         | 3/32 [00:00<00:09,  3.00it/s]
 12%|█▎        | 4/32 [00:01<00:08,  3.21it/s]
 16%|█▌        | 5/32 [00:01<00:08,  3.35it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.38it/s]
 22%|██▏       | 7/32 [00:02<00:07,  3.46it/s]
 25%|██▌       | 8/32 [00:02<00:06,  3.45it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.48it/s]
 31%|███▏      | 10/32 [00:02<00:06,  3.59it/s]
 34%|███▍      | 11/32 [00:03<00:05,  3.62it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.60it/s]
 41%|████      | 13/32 [00:03<00:05,  3.58it/s]
 44%|████▍     | 14/32 [00:04<00:05,  3.49it/s]
 47%|████▋     | 15/32 [00:04<00:04,  3.55it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.58it/s]
 53%|█████▎    | 17/32 [00:04<00:04,  3.65it/s]
 56%|█████▋    | 18/32 [00:05<00:03,  3.66it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.66it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.64it/s]
 66%|██████▌   | 21/32 [00:05<00:03,  3.62it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.69it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.70it/s]
 75%|███████▌  | 24/32 [00:06<00:02,  3.66it/s]
 78%|███████▊  | 25/32 [00:07<00:01,  3.60it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.63it/s]
 84%|████████▍ | 27/32 [00:07<00:01,  3.60it/s]
 88%|████████▊ | 28/32 [00:07<00:01,  3.65it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.67it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.66it/s]
 97%|█████████▋| 31/32 [00:08<00:00,  3.59it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:06,  2.87it/s]
 11%|█         | 2/19 [00:00<00:05,  3.26it/s]
 16%|█▌        | 3/19 [00:00<00:05,  3.08it/s]
 21%|██        | 4/19 [00:01<00:04,  3.24it/s]
 26%|██▋       | 5/19 [00:01<00:04,  3.31it/s]
 32%|███▏      | 6/19 [00:01<00:03,  3.45it/s]
 37%|███▋      | 7/19 [00:02<00:03,  3.51it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.47it/s]
 47%|████▋     | 9/19 [00:02<00:02,  3.50it/s]
 53%|█████▎    | 10/19 [00:02<00:02,  3.43it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.44it/s]
 63%|██████▎   | 12/19 [00:03<00:02,  3.48it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.49it/s]
 74%|███████▎  | 14/19 [00:04<00:01,  3.51it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.58it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.59it/s]
 89%|████████▉ | 17/19 [00:04<00:00,  3.62it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.59it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:08,  3.16it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.42it/s]
 11%|█         | 3/27 [00:00<00:06,  3.58it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.56it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.60it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.64it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.65it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.65it/s]
 33%|███▎      | 9/27 [00:02<00:04,  3.68it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.67it/s]
 41%|████      | 11/27 [00:03<00:04,  3.65it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.62it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.68it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.66it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.71it/s]
 59%|█████▉    | 16/27 [00:04<00:02,  3.74it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.67it/s]
 67%|██████▋   | 18/27 [00:04<00:02,  3.60it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.59it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.66it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.62it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.67it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.67it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.69it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.65it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.66it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:04,  2.90it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.09it/s]
 23%|██▎       | 3/13 [00:00<00:03,  3.27it/s]
 31%|███       | 4/13 [00:01<00:02,  3.40it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.50it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.56it/s]
 54%|█████▍    | 7/13 [00:02<00:01,  3.56it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.55it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.56it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.55it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.54it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.61it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.50it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.55it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:01,  3.61it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.50it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.58it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.51it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.49it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.54it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.56it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.46it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.55it/s]
 12%|█▎        | 3/24 [00:00<00:06,  3.49it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.49it/s]
 21%|██        | 5/24 [00:01<00:05,  3.48it/s]
 25%|██▌       | 6/24 [00:01<00:05,  3.45it/s]
 29%|██▉       | 7/24 [00:02<00:04,  3.52it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.57it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.59it/s]
 42%|████▏     | 10/24 [00:02<00:03,  3.57it/s]
 46%|████▌     | 11/24 [00:03<00:03,  3.59it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.60it/s]
 54%|█████▍    | 13/24 [00:03<00:03,  3.61it/s]
 58%|█████▊    | 14/24 [00:03<00:02,  3.61it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.64it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.65it/s]
 71%|███████   | 17/24 [00:04<00:01,  3.65it/s]
 75%|███████▌  | 18/24 [00:05<00:01,  3.64it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.63it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.68it/s]
 88%|████████▊ | 21/24 [00:05<00:00,  3.68it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.65it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.64it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.66it/s]
 22%|██▏       | 2/9 [00:00<00:01,  3.67it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.61it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.58it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.62it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.67it/s]
 78%|███████▊  | 7/9 [00:01<00:00,  3.76it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.74it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.56it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.66it/s]
 11%|█         | 3/27 [00:00<00:06,  3.70it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.69it/s]
 19%|█▊        | 5/27 [00:01<00:05,  3.68it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.75it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.68it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.68it/s]
 33%|███▎      | 9/27 [00:02<00:04,  3.70it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.72it/s]
 41%|████      | 11/27 [00:02<00:04,  3.70it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.67it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.68it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.67it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.64it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.65it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.65it/s]
 67%|██████▋   | 18/27 [00:04<00:02,  3.65it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.59it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.59it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.58it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.55it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.56it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.58it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.56it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.59it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  3.54it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.53it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.52it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.51it/s]
 42%|████▏     | 5/12 [00:01<00:01,  3.61it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.66it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.73it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.71it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.73it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.71it/s]
 92%|█████████▏| 11/12 [00:03<00:00,  3.73it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.64it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.67it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.66it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.71it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.76it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.68it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.68it/s]
 15%|█▌        | 2/13 [00:00<00:02,  3.68it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.64it/s]
 31%|███       | 4/13 [00:01<00:02,  3.60it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.63it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.61it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.62it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.64it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.63it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.62it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.63it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.59it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.69it/s]
 10%|█         | 2/20 [00:00<00:04,  3.71it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.69it/s]
 20%|██        | 4/20 [00:01<00:04,  3.71it/s]
 25%|██▌       | 5/20 [00:01<00:03,  3.76it/s]
 30%|███       | 6/20 [00:01<00:03,  3.72it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.67it/s]
 40%|████      | 8/20 [00:02<00:03,  3.72it/s]
 45%|████▌     | 9/20 [00:02<00:02,  3.73it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.73it/s]
 55%|█████▌    | 11/20 [00:02<00:02,  3.75it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.73it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.78it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.79it/s]
 75%|███████▌  | 15/20 [00:04<00:01,  3.76it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.83it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.76it/s]
 90%|█████████ | 18/20 [00:04<00:00,  3.76it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.73it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:03,  3.70it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.85it/s]
 20%|██        | 3/15 [00:00<00:03,  3.78it/s]
 27%|██▋       | 4/15 [00:01<00:02,  3.74it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.70it/s]
 40%|████      | 6/15 [00:01<00:02,  3.81it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.76it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.69it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.68it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.67it/s]
 73%|███████▎  | 11/15 [00:02<00:01,  3.69it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.73it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.76it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.76it/s]
+
+[2025-07-11 17:08:03] Weighted Per image PCK0.10: 47.56%, image PCK0.05: 31.36%, image PCK0.01: 57.68%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.3_layer_18_no_dino.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:06:46] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.3_layer_18_no_dino', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.3_layer_18'], NO_DINO=True, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='7', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.3_layer_18']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:20,  1.44s/it]
 13%|█▎        | 2/15 [00:01<00:11,  1.09it/s]
 20%|██        | 3/15 [00:02<00:07,  1.58it/s]
 27%|██▋       | 4/15 [00:02<00:05,  1.99it/s]
 33%|███▎      | 5/15 [00:02<00:04,  2.37it/s]
 40%|████      | 6/15 [00:03<00:03,  2.63it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.92it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  3.06it/s]
 60%|██████    | 9/15 [00:04<00:01,  3.18it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.28it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.38it/s]
 80%|████████  | 12/15 [00:04<00:00,  3.55it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.66it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.71it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:08,  3.47it/s]
  7%|▋         | 2/30 [00:00<00:07,  3.58it/s]
 10%|█         | 3/30 [00:00<00:07,  3.55it/s]
 13%|█▎        | 4/30 [00:01<00:07,  3.58it/s]
 17%|█▋        | 5/30 [00:01<00:06,  3.67it/s]
 20%|██        | 6/30 [00:01<00:06,  3.70it/s]
 23%|██▎       | 7/30 [00:01<00:06,  3.66it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.58it/s]
 30%|███       | 9/30 [00:02<00:05,  3.55it/s]
 33%|███▎      | 10/30 [00:02<00:05,  3.54it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.59it/s]
 40%|████      | 12/30 [00:03<00:05,  3.58it/s]
 43%|████▎     | 13/30 [00:03<00:04,  3.62it/s]
 47%|████▋     | 14/30 [00:03<00:04,  3.56it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.57it/s]
 53%|█████▎    | 16/30 [00:04<00:03,  3.58it/s]
 57%|█████▋    | 17/30 [00:04<00:03,  3.64it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.62it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.62it/s]
 67%|██████▋   | 20/30 [00:05<00:02,  3.59it/s]
 70%|███████   | 21/30 [00:05<00:02,  3.56it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.53it/s]
 77%|███████▋  | 23/30 [00:06<00:01,  3.62it/s]
 80%|████████  | 24/30 [00:06<00:01,  3.56it/s]
 83%|████████▎ | 25/30 [00:06<00:01,  3.57it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.59it/s]
 90%|█████████ | 27/30 [00:07<00:00,  3.55it/s]
 93%|█████████▎| 28/30 [00:07<00:00,  3.57it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.57it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.49it/s]
 20%|██        | 2/10 [00:00<00:02,  3.55it/s]
 30%|███       | 3/10 [00:00<00:01,  3.68it/s]
 40%|████      | 4/10 [00:01<00:01,  3.66it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.64it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.71it/s]
 70%|███████   | 7/10 [00:01<00:00,  3.70it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.70it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.61it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.51it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.52it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.63it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.70it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.76it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:01,  3.72it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.68it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.71it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.71it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.71it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.70it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.62it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:09,  3.44it/s]
  6%|▋         | 2/32 [00:00<00:08,  3.52it/s]
  9%|▉         | 3/32 [00:00<00:08,  3.55it/s]
 12%|█▎        | 4/32 [00:01<00:07,  3.62it/s]
 16%|█▌        | 5/32 [00:01<00:07,  3.62it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.69it/s]
 22%|██▏       | 7/32 [00:01<00:06,  3.69it/s]
 25%|██▌       | 8/32 [00:02<00:06,  3.73it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.71it/s]
 31%|███▏      | 10/32 [00:02<00:06,  3.64it/s]
 34%|███▍      | 11/32 [00:03<00:05,  3.62it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.60it/s]
 41%|████      | 13/32 [00:03<00:05,  3.63it/s]
 44%|████▍     | 14/32 [00:03<00:04,  3.67it/s]
 47%|████▋     | 15/32 [00:04<00:04,  3.67it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.66it/s]
 53%|█████▎    | 17/32 [00:04<00:04,  3.63it/s]
 56%|█████▋    | 18/32 [00:04<00:03,  3.63it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.63it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.64it/s]
 66%|██████▌   | 21/32 [00:05<00:03,  3.66it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.59it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.57it/s]
 75%|███████▌  | 24/32 [00:06<00:02,  3.62it/s]
 78%|███████▊  | 25/32 [00:06<00:01,  3.66it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.61it/s]
 84%|████████▍ | 27/32 [00:07<00:01,  3.62it/s]
 88%|████████▊ | 28/32 [00:07<00:01,  3.57it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.55it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.54it/s]
 97%|█████████▋| 31/32 [00:08<00:00,  3.61it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:05,  3.55it/s]
 11%|█         | 2/19 [00:00<00:04,  3.58it/s]
 16%|█▌        | 3/19 [00:00<00:04,  3.60it/s]
 21%|██        | 4/19 [00:01<00:04,  3.61it/s]
 26%|██▋       | 5/19 [00:01<00:03,  3.63it/s]
 32%|███▏      | 6/19 [00:01<00:03,  3.57it/s]
 37%|███▋      | 7/19 [00:01<00:03,  3.56it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.62it/s]
 47%|████▋     | 9/19 [00:02<00:02,  3.61it/s]
 53%|█████▎    | 10/19 [00:02<00:02,  3.65it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.64it/s]
 63%|██████▎   | 12/19 [00:03<00:01,  3.56it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.59it/s]
 74%|███████▎  | 14/19 [00:03<00:01,  3.63it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.60it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.60it/s]
 89%|████████▉ | 17/19 [00:04<00:00,  3.60it/s]
 95%|█████████▍| 18/19 [00:04<00:00,  3.65it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.67it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.61it/s]
 11%|█         | 3/27 [00:00<00:06,  3.58it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.66it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.65it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.61it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.65it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.56it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.58it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.63it/s]
 41%|████      | 11/27 [00:03<00:04,  3.66it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.71it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.64it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.65it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.63it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.65it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.66it/s]
 67%|██████▋   | 18/27 [00:04<00:02,  3.71it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.72it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.68it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.64it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.62it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.62it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.61it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.63it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.63it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.37it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.59it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.60it/s]
 31%|███       | 4/13 [00:01<00:02,  3.60it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.61it/s]
 46%|████▌     | 6/13 [00:01<00:02,  3.46it/s]
 54%|█████▍    | 7/13 [00:02<00:01,  3.43it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.51it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.56it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.60it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.64it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.65it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.70it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.72it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:01,  3.57it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.63it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.66it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.72it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.71it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.68it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.70it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.61it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.66it/s]
 12%|█▎        | 3/24 [00:00<00:05,  3.68it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.75it/s]
 21%|██        | 5/24 [00:01<00:05,  3.73it/s]
 25%|██▌       | 6/24 [00:01<00:04,  3.72it/s]
 29%|██▉       | 7/24 [00:01<00:04,  3.71it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.70it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.69it/s]
 42%|████▏     | 10/24 [00:02<00:03,  3.72it/s]
 46%|████▌     | 11/24 [00:02<00:03,  3.72it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.71it/s]
 54%|█████▍    | 13/24 [00:03<00:02,  3.71it/s]
 58%|█████▊    | 14/24 [00:03<00:02,  3.67it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.67it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.69it/s]
 71%|███████   | 17/24 [00:04<00:01,  3.68it/s]
 75%|███████▌  | 18/24 [00:04<00:01,  3.70it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.71it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.68it/s]
 88%|████████▊ | 21/24 [00:05<00:00,  3.68it/s]
 92%|█████████▏| 22/24 [00:05<00:00,  3.70it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.71it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.70it/s]
 22%|██▏       | 2/9 [00:00<00:01,  3.75it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.78it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.84it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.75it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.71it/s]
 78%|███████▊  | 7/9 [00:01<00:00,  3.67it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.68it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:06,  3.75it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.71it/s]
 11%|█         | 3/27 [00:00<00:06,  3.68it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.63it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.65it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.61it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.67it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.67it/s]
 33%|███▎      | 9/27 [00:02<00:04,  3.68it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.66it/s]
 41%|████      | 11/27 [00:03<00:04,  3.66it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.69it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.74it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.75it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.76it/s]
 59%|█████▉    | 16/27 [00:04<00:02,  3.73it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.72it/s]
 67%|██████▋   | 18/27 [00:04<00:02,  3.73it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.75it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.72it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.72it/s]
 81%|████████▏ | 22/27 [00:05<00:01,  3.79it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.75it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.79it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.85it/s]
 96%|█████████▋| 26/27 [00:06<00:00,  3.82it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:02,  3.70it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.89it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.94it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.97it/s]
 42%|████▏     | 5/12 [00:01<00:01,  3.85it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.85it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.80it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.80it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.76it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.76it/s]
 92%|█████████▏| 11/12 [00:02<00:00,  3.74it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.56it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.59it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.64it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.64it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.60it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.64it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.69it/s]
 15%|█▌        | 2/13 [00:00<00:02,  3.74it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.75it/s]
 31%|███       | 4/13 [00:01<00:02,  3.77it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.72it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.71it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.71it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.67it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.65it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.72it/s]
 85%|████████▍ | 11/13 [00:02<00:00,  3.66it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.74it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.62it/s]
 10%|█         | 2/20 [00:00<00:04,  3.69it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.74it/s]
 20%|██        | 4/20 [00:01<00:04,  3.76it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.73it/s]
 30%|███       | 6/20 [00:01<00:03,  3.76it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.82it/s]
 40%|████      | 8/20 [00:02<00:03,  3.75it/s]
 45%|████▌     | 9/20 [00:02<00:02,  3.72it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.74it/s]
 55%|█████▌    | 11/20 [00:02<00:02,  3.73it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.76it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.71it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.70it/s]
 75%|███████▌  | 15/20 [00:04<00:01,  3.72it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.69it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.79it/s]
 90%|█████████ | 18/20 [00:04<00:00,  3.74it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.77it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:03,  3.57it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.61it/s]
 20%|██        | 3/15 [00:00<00:03,  3.66it/s]
 27%|██▋       | 4/15 [00:01<00:03,  3.64it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.69it/s]
 40%|████      | 6/15 [00:01<00:02,  3.64it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.69it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.72it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.73it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.74it/s]
 73%|███████▎  | 11/15 [00:02<00:01,  3.77it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.78it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.75it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.72it/s]
+
+[2025-07-11 17:08:10] Weighted Per image PCK0.10: 32.57%, image PCK0.05: 19.31%, image PCK0.01: 41.81%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.3_layer_5_no_dino.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:01:56] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.3_layer_5_no_dino', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.3_layer_5'], NO_DINO=True, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='5', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.3_layer_5']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:22,  1.62s/it]
 13%|█▎        | 2/15 [00:02<00:12,  1.01it/s]
 20%|██        | 3/15 [00:02<00:08,  1.48it/s]
 27%|██▋       | 4/15 [00:02<00:05,  1.87it/s]
 33%|███▎      | 5/15 [00:03<00:04,  2.23it/s]
 40%|████      | 6/15 [00:03<00:03,  2.57it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.78it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  2.92it/s]
 60%|██████    | 9/15 [00:04<00:01,  3.02it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.13it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.04it/s]
 80%|████████  | 12/15 [00:05<00:00,  3.14it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.23it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.32it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:09,  3.17it/s]
  7%|▋         | 2/30 [00:00<00:08,  3.30it/s]
 10%|█         | 3/30 [00:00<00:08,  3.27it/s]
 13%|█▎        | 4/30 [00:01<00:07,  3.29it/s]
 17%|█▋        | 5/30 [00:01<00:07,  3.32it/s]
 20%|██        | 6/30 [00:01<00:07,  3.19it/s]
 23%|██▎       | 7/30 [00:02<00:07,  3.26it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.35it/s]
 30%|███       | 9/30 [00:02<00:06,  3.31it/s]
 33%|███▎      | 10/30 [00:03<00:06,  3.05it/s]
 37%|███▋      | 11/30 [00:03<00:06,  3.13it/s]
 40%|████      | 12/30 [00:03<00:05,  3.20it/s]
 43%|████▎     | 13/30 [00:04<00:05,  3.23it/s]
 47%|████▋     | 14/30 [00:04<00:04,  3.33it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.27it/s]
 53%|█████▎    | 16/30 [00:04<00:04,  3.09it/s]
 57%|█████▋    | 17/30 [00:05<00:04,  3.16it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.18it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.25it/s]
 67%|██████▋   | 20/30 [00:06<00:03,  3.27it/s]
 70%|███████   | 21/30 [00:06<00:02,  3.33it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.36it/s]
 77%|███████▋  | 23/30 [00:07<00:02,  3.35it/s]
 80%|████████  | 24/30 [00:07<00:01,  3.37it/s]
 83%|████████▎ | 25/30 [00:07<00:01,  3.37it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.35it/s]
 90%|█████████ | 27/30 [00:08<00:00,  3.37it/s]
 93%|█████████▎| 28/30 [00:08<00:00,  3.37it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.33it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.34it/s]
 20%|██        | 2/10 [00:00<00:02,  3.39it/s]
 30%|███       | 3/10 [00:00<00:02,  3.20it/s]
 40%|████      | 4/10 [00:01<00:01,  3.20it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.22it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.24it/s]
 70%|███████   | 7/10 [00:02<00:00,  3.32it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.40it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.40it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  2.86it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.24it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.22it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.25it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.32it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.35it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.49it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.42it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.39it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.43it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.42it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.44it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:09,  3.35it/s]
  6%|▋         | 2/32 [00:00<00:08,  3.42it/s]
  9%|▉         | 3/32 [00:00<00:08,  3.38it/s]
 12%|█▎        | 4/32 [00:01<00:08,  3.41it/s]
 16%|█▌        | 5/32 [00:01<00:07,  3.44it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.38it/s]
 22%|██▏       | 7/32 [00:02<00:07,  3.43it/s]
 25%|██▌       | 8/32 [00:02<00:07,  3.40it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.41it/s]
 31%|███▏      | 10/32 [00:02<00:06,  3.44it/s]
 34%|███▍      | 11/32 [00:03<00:06,  3.46it/s]
 38%|███▊      | 12/32 [00:03<00:06,  3.28it/s]
 41%|████      | 13/32 [00:03<00:05,  3.32it/s]
 44%|████▍     | 14/32 [00:04<00:05,  3.33it/s]
 47%|████▋     | 15/32 [00:04<00:05,  3.38it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.42it/s]
 53%|█████▎    | 17/32 [00:04<00:04,  3.48it/s]
 56%|█████▋    | 18/32 [00:05<00:04,  3.45it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.48it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.45it/s]
 66%|██████▌   | 21/32 [00:06<00:03,  3.24it/s]
 69%|██████▉   | 22/32 [00:06<00:03,  3.18it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.26it/s]
 75%|███████▌  | 24/32 [00:07<00:02,  3.26it/s]
 78%|███████▊  | 25/32 [00:07<00:02,  3.27it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.32it/s]
 84%|████████▍ | 27/32 [00:08<00:01,  3.32it/s]
 88%|████████▊ | 28/32 [00:09<00:03,  1.32it/s]
 91%|█████████ | 29/32 [00:10<00:01,  1.61it/s]
 94%|█████████▍| 30/32 [00:10<00:01,  1.89it/s]
 97%|█████████▋| 31/32 [00:10<00:00,  2.10it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:06,  2.67it/s]
 11%|█         | 2/19 [00:00<00:05,  2.86it/s]
 16%|█▌        | 3/19 [00:01<00:05,  3.03it/s]
 21%|██        | 4/19 [00:01<00:04,  3.13it/s]
 26%|██▋       | 5/19 [00:01<00:04,  3.17it/s]
 32%|███▏      | 6/19 [00:01<00:03,  3.26it/s]
 37%|███▋      | 7/19 [00:02<00:03,  3.31it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.30it/s]
 47%|████▋     | 9/19 [00:02<00:03,  3.25it/s]
 53%|█████▎    | 10/19 [00:03<00:02,  3.24it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.26it/s]
 63%|██████▎   | 12/19 [00:03<00:02,  3.31it/s]
 68%|██████▊   | 13/19 [00:04<00:01,  3.27it/s]
 74%|███████▎  | 14/19 [00:04<00:01,  3.27it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.32it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.37it/s]
 89%|████████▉ | 17/19 [00:05<00:00,  3.38it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.39it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:10,  2.44it/s]
  7%|▋         | 2/27 [00:00<00:08,  2.87it/s]
 11%|█         | 3/27 [00:01<00:07,  3.04it/s]
 15%|█▍        | 4/27 [00:01<00:07,  3.13it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.23it/s]
 22%|██▏       | 6/27 [00:01<00:06,  3.27it/s]
 26%|██▌       | 7/27 [00:02<00:06,  3.26it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.29it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.37it/s]
 37%|███▋      | 10/27 [00:03<00:05,  3.39it/s]
 41%|████      | 11/27 [00:03<00:04,  3.36it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.34it/s]
 48%|████▊     | 13/27 [00:03<00:04,  3.40it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.40it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.47it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.50it/s]
 63%|██████▎   | 17/27 [00:05<00:02,  3.46it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.42it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.43it/s]
 74%|███████▍  | 20/27 [00:06<00:02,  3.42it/s]
 78%|███████▊  | 21/27 [00:06<00:01,  3.41it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.44it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.42it/s]
 89%|████████▉ | 24/27 [00:07<00:00,  3.44it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.40it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.39it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.04it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.16it/s]
 23%|██▎       | 3/13 [00:00<00:03,  3.17it/s]
 31%|███       | 4/13 [00:01<00:02,  3.25it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.26it/s]
 46%|████▌     | 6/13 [00:01<00:02,  3.31it/s]
 54%|█████▍    | 7/13 [00:02<00:01,  3.31it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.29it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.31it/s]
 77%|███████▋  | 10/13 [00:03<00:00,  3.32it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.35it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.40it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.34it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.41it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.37it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.36it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.44it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.44it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.50it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.48it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.54it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:07,  3.21it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.27it/s]
 12%|█▎        | 3/24 [00:00<00:06,  3.36it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.35it/s]
 21%|██        | 5/24 [00:01<00:05,  3.35it/s]
 25%|██▌       | 6/24 [00:01<00:05,  3.39it/s]
 29%|██▉       | 7/24 [00:02<00:04,  3.42it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.41it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.36it/s]
 42%|████▏     | 10/24 [00:02<00:04,  3.34it/s]
 46%|████▌     | 11/24 [00:03<00:03,  3.30it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.33it/s]
 54%|█████▍    | 13/24 [00:03<00:03,  3.35it/s]
 58%|█████▊    | 14/24 [00:04<00:02,  3.40it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.49it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.51it/s]
 71%|███████   | 17/24 [00:04<00:01,  3.52it/s]
 75%|███████▌  | 18/24 [00:05<00:01,  3.53it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.52it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.57it/s]
 88%|████████▊ | 21/24 [00:06<00:00,  3.55it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.55it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.53it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.03it/s]
 22%|██▏       | 2/9 [00:00<00:02,  3.24it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.16it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.23it/s]
 56%|█████��    | 5/9 [00:01<00:01,  3.32it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.41it/s]
 78%|███████▊  | 7/9 [00:02<00:00,  3.52it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.52it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.49it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.52it/s]
 11%|█         | 3/27 [00:00<00:06,  3.57it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.56it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.54it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.63it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.57it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.58it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.58it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.59it/s]
 41%|████      | 11/27 [00:03<00:04,  3.57it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.56it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.55it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.57it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.55it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.55it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.56it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.56it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.53it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.56it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.53it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.49it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.49it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.48it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.47it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.50it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  2.95it/s]
 17%|█▋        | 2/12 [00:00<00:03,  3.21it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.25it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.31it/s]
 42%|████▏     | 5/12 [00:01<00:02,  3.45it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.50it/s]
 58%|█████▊    | 7/12 [00:02<00:01,  3.57it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.57it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.62it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.59it/s]
 92%|█████████▏| 11/12 [00:03<00:00,  3.62it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.32it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.53it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.57it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.63it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.69it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.61it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.41it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.44it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.51it/s]
 31%|███       | 4/13 [00:01<00:02,  3.52it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.55it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.54it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.52it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.55it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.60it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.55it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.62it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.56it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.59it/s]
 10%|█         | 2/20 [00:00<00:05,  3.59it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.59it/s]
 20%|██        | 4/20 [00:01<00:04,  3.57it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.65it/s]
 30%|███       | 6/20 [00:01<00:03,  3.61it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.63it/s]
 40%|████      | 8/20 [00:02<00:03,  3.69it/s]
 45%|████▌     | 9/20 [00:02<00:02,  3.74it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.70it/s]
 55%|█████▌    | 11/20 [00:02<00:02,  3.73it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.65it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.67it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.71it/s]
 75%|███████▌  | 15/20 [00:04<00:01,  3.71it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.43it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.41it/s]
 90%|█████████ | 18/20 [00:04<00:00,  3.52it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.52it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:03,  3.58it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.73it/s]
 20%|██        | 3/15 [00:00<00:03,  3.62it/s]
 27%|██▋       | 4/15 [00:01<00:03,  3.66it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.67it/s]
 40%|████      | 6/15 [00:01<00:02,  3.74it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.68it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.62it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.60it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.58it/s]
 73%|███████▎  | 11/15 [00:03<00:01,  3.59it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.59it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.58it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.63it/s]
+
+[2025-07-11 17:03:28] Weighted Per image PCK0.10: 43.51%, image PCK0.05: 20.05%, image PCK0.01: 61.65%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.3_layer_9.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:01:56] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.3_layer_9', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.3_layer_9'], NO_DINO=False, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='6', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.3_layer_9']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:21,  1.55s/it]
 13%|█▎        | 2/15 [00:02<00:12,  1.01it/s]
 20%|██        | 3/15 [00:02<00:08,  1.47it/s]
 27%|██▋       | 4/15 [00:02<00:05,  1.87it/s]
 33%|███▎      | 5/15 [00:03<00:04,  2.20it/s]
 40%|████      | 6/15 [00:03<00:03,  2.48it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.67it/s]
 53%|█████▎    | 8/15 [00:04<00:02,  2.83it/s]
 60%|██████    | 9/15 [00:04<00:02,  2.98it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.12it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.18it/s]
 80%|████████  | 12/15 [00:05<00:00,  3.25it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.31it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.33it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:08,  3.37it/s]
  7%|▋         | 2/30 [00:00<00:10,  2.55it/s]
 10%|█         | 3/30 [00:01<00:09,  2.89it/s]
 13%|█▎        | 4/30 [00:01<00:08,  3.00it/s]
 17%|█▋        | 5/30 [00:01<00:08,  3.03it/s]
 20%|██        | 6/30 [00:01<00:07,  3.14it/s]
 23%|██▎       | 7/30 [00:02<00:07,  3.26it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.37it/s]
 30%|███       | 9/30 [00:02<00:06,  3.41it/s]
 33%|███▎      | 10/30 [00:03<00:05,  3.48it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.46it/s]
 40%|████      | 12/30 [00:03<00:05,  3.44it/s]
 43%|████▎     | 13/30 [00:03<00:04,  3.44it/s]
 47%|████▋     | 14/30 [00:04<00:04,  3.54it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.28it/s]
 53%|█████▎    | 16/30 [00:04<00:04,  3.27it/s]
 57%|█████▋    | 17/30 [00:05<00:03,  3.30it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.34it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.38it/s]
 67%|██████▋   | 20/30 [00:06<00:02,  3.42it/s]
 70%|███████   | 21/30 [00:06<00:02,  3.49it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.53it/s]
 77%|███████▋  | 23/30 [00:06<00:02,  3.24it/s]
 80%|████████  | 24/30 [00:07<00:01,  3.20it/s]
 83%|████████▎ | 25/30 [00:07<00:01,  3.19it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.09it/s]
 90%|█████████ | 27/30 [00:08<00:00,  3.16it/s]
 93%|█████████▎| 28/30 [00:08<00:00,  3.18it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.23it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.23it/s]
 20%|██        | 2/10 [00:00<00:02,  3.08it/s]
 30%|███       | 3/10 [00:00<00:02,  3.14it/s]
 40%|████      | 4/10 [00:01<00:01,  3.22it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.29it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.28it/s]
 70%|███████   | 7/10 [00:02<00:00,  3.35it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.37it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.37it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.60it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.56it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.50it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.45it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.44it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.37it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.49it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.46it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.45it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.45it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.44it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.51it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:08,  3.62it/s]
  6%|▋         | 2/32 [00:00<00:08,  3.47it/s]
  9%|▉         | 3/32 [00:00<00:08,  3.48it/s]
 12%|█▎        | 4/32 [00:01<00:08,  3.49it/s]
 16%|█▌        | 5/32 [00:01<00:07,  3.52it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.45it/s]
 22%|██▏       | 7/32 [00:02<00:07,  3.47it/s]
 25%|██▌       | 8/32 [00:02<00:07,  3.43it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.35it/s]
 31%|███▏      | 10/32 [00:02<00:06,  3.46it/s]
 34%|███▍      | 11/32 [00:03<00:05,  3.52it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.50it/s]
 41%|████      | 13/32 [00:03<00:05,  3.47it/s]
 44%|████▍     | 14/32 [00:04<00:05,  3.47it/s]
 47%|████▋     | 15/32 [00:04<00:05,  3.27it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.29it/s]
 53%|█████▎    | 17/32 [00:04<00:04,  3.36it/s]
 56%|█████▋    | 18/32 [00:05<00:04,  3.31it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.39it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.41it/s]
 66%|██████▌   | 21/32 [00:06<00:03,  3.43it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.52it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.44it/s]
 75%|███████▌  | 24/32 [00:06<00:02,  3.43it/s]
 78%|███████▊  | 25/32 [00:07<00:02,  3.42it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.55it/s]
 84%|████████▍ | 27/32 [00:07<00:01,  3.52it/s]
 88%|████████▊ | 28/32 [00:10<00:03,  1.08it/s]
 91%|█████████ | 29/32 [00:10<00:02,  1.35it/s]
 94%|█████████▍| 30/32 [00:10<00:01,  1.63it/s]
 97%|█████████▋| 31/32 [00:11<00:00,  1.88it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:05,  3.37it/s]
 11%|█         | 2/19 [00:00<00:04,  3.52it/s]
 16%|█▌        | 3/19 [00:00<00:04,  3.56it/s]
 21%|██        | 4/19 [00:01<00:04,  3.54it/s]
 26%|██▋       | 5/19 [00:01<00:03,  3.52it/s]
 32%|███▏      | 6/19 [00:01<00:03,  3.59it/s]
 37%|███▋      | 7/19 [00:01<00:03,  3.66it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.61it/s]
 47%|████▋     | 9/19 [00:02<00:02,  3.62it/s]
 53%|█████▎    | 10/19 [00:02<00:02,  3.54it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.53it/s]
 63%|██████▎   | 12/19 [00:03<00:01,  3.58it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.55it/s]
 74%|███████▎  | 14/19 [00:03<00:01,  3.55it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.61it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.64it/s]
 89%|████████▉ | 17/19 [00:04<00:00,  3.67it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.62it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.46it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.59it/s]
 11%|█         | 3/27 [00:00<00:06,  3.63it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.55it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.59it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.62it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.61it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.60it/s]
 33%|███▎      | 9/27 [00:02<00:04,  3.62it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.61it/s]
 41%|████      | 11/27 [00:03<00:04,  3.59it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.55it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.59it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.57it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.60it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.64it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.57it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.51it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.52it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.56it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.54it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.58it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.53it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.52it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.52it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.48it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.55it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.46it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.46it/s]
 31%|███       | 4/13 [00:01<00:02,  3.51it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.49it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.54it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.53it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.48it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.40it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.33it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.38it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.45it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.38it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.48it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.33it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.36it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.41it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.43it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.41it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.48it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.48it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.46it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.47it/s]
 12%|█▎        | 3/24 [00:00<00:05,  3.51it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.45it/s]
 21%|██        | 5/24 [00:01<00:05,  3.43it/s]
 25%|██▌       | 6/24 [00:01<00:05,  3.47it/s]
 29%|██▉       | 7/24 [00:02<00:04,  3.50it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.52it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.52it/s]
 42%|████▏     | 10/24 [00:02<00:04,  3.47it/s]
 46%|████▌     | 11/24 [00:03<00:03,  3.45it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.45it/s]
 54%|█████▍    | 13/24 [00:03<00:03,  3.46it/s]
 58%|█████▊    | 14/24 [00:04<00:02,  3.41it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.47it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.47it/s]
 71%|███████   | 17/24 [00:04<00:02,  3.46it/s]
 75%|███████▌  | 18/24 [00:05<00:01,  3.44it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.45it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.51it/s]
 88%|████████▊ | 21/24 [00:06<00:00,  3.49it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.48it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.47it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.54it/s]
 22%|██▏       | 2/9 [00:00<00:01,  3.55it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.49it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.41it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.41it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.47it/s]
 78%|███████▊  | 7/9 [00:02<00:00,  3.54it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.54it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.44it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.46it/s]
 11%|█         | 3/27 [00:00<00:06,  3.49it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.47it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.42it/s]
 22%|██▏       | 6/27 [00:01<00:06,  3.47it/s]
 26%|██▌       | 7/27 [00:02<00:05,  3.43it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.42it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.45it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.46it/s]
 41%|████      | 11/27 [00:03<00:04,  3.44it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.43it/s]
 48%|████▊     | 13/27 [00:03<00:04,  3.42it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.42it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.42it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.43it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.47it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.50it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.49it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.52it/s]
 78%|███████▊  | 21/27 [00:06<00:01,  3.54it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.50it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.52it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.52it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.48it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.51it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  3.57it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.51it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.48it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.45it/s]
 42%|████▏     | 5/12 [00:01<00:01,  3.52it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.54it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.60it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.58it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.60it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.59it/s]
 92%|█████████▏| 11/12 [00:03<00:00,  3.60it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.54it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.60it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.53it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.54it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.58it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.57it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.57it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.55it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.51it/s]
 31%|███       | 4/13 [00:01<00:02,  3.48it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.50it/s]
 46%|████▌     | 6/13 [00:01<00:02,  3.47it/s]
 54%|█████▍    | 7/13 [00:02<00:01,  3.46it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.49it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.53it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.50it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.55it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.49it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.22it/s]
 10%|█         | 2/20 [00:00<00:05,  3.42it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.48it/s]
 20%|██        | 4/20 [00:01<00:04,  3.52it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.57it/s]
 30%|███       | 6/20 [00:01<00:03,  3.55it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.54it/s]
 40%|████      | 8/20 [00:02<00:03,  3.61it/s]
 45%|████▌     | 9/20 [00:02<00:03,  3.63it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.59it/s]
 55%|█████▌    | 11/20 [00:03<00:02,  3.64it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.60it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.61it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.57it/s]
 75%|███████▌  | 15/20 [00:04<00:01,  3.50it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.53it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.44it/s]
 90%|█████████ | 18/20 [00:05<00:00,  3.47it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.44it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:04,  3.42it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.53it/s]
 20%|██        | 3/15 [00:00<00:03,  3.48it/s]
 27%|██▋       | 4/15 [00:01<00:03,  3.51it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.47it/s]
 40%|████      | 6/15 [00:01<00:02,  3.52it/s]
 47%|████▋     | 7/15 [00:02<00:02,  3.46it/s]
 53%|█████▎    | 8/15 [00:02<00:02,  3.43it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.42it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.39it/s]
 73%|███████▎  | 11/15 [00:03<00:01,  3.41it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.41it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.44it/s]
 93%|█████████▎| 14/15 [00:04<00:00,  3.47it/s]
+
+[2025-07-11 17:03:27] Weighted Per image PCK0.10: 74.31%, image PCK0.05: 56.82%, image PCK0.01: 82.30%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.5_layer_10.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:09:46] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.5_layer_10', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.5_layer_10'], NO_DINO=False, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='6', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.5_layer_10']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:20,  1.47s/it]
 13%|█▎        | 2/15 [00:02<00:11,  1.08it/s]
 20%|██        | 3/15 [00:02<00:07,  1.51it/s]
 27%|██▋       | 4/15 [00:02<00:05,  1.94it/s]
 33%|███▎      | 5/15 [00:02<00:04,  2.32it/s]
 40%|████      | 6/15 [00:03<00:03,  2.68it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.86it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  3.07it/s]
 60%|██████    | 9/15 [00:04<00:01,  3.11it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.21it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.26it/s]
 80%|████████  | 12/15 [00:04<00:00,  3.28it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.35it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.41it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:08,  3.35it/s]
  7%|▋         | 2/30 [00:00<00:08,  3.32it/s]
 10%|█         | 3/30 [00:00<00:07,  3.42it/s]
 13%|█▎        | 4/30 [00:01<00:07,  3.39it/s]
 17%|█▋        | 5/30 [00:01<00:07,  3.33it/s]
 20%|██        | 6/30 [00:01<00:07,  3.31it/s]
 23%|██▎       | 7/30 [00:02<00:06,  3.37it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.45it/s]
 30%|███       | 9/30 [00:02<00:06,  3.41it/s]
 33%|███▎      | 10/30 [00:02<00:05,  3.41it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.36it/s]
 40%|████      | 12/30 [00:03<00:05,  3.38it/s]
 43%|████▎     | 13/30 [00:03<00:05,  3.37it/s]
 47%|████▋     | 14/30 [00:04<00:04,  3.43it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.41it/s]
 53%|█████▎    | 16/30 [00:04<00:04,  3.39it/s]
 57%|█████▋    | 17/30 [00:05<00:03,  3.39it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.37it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.38it/s]
 67%|██████▋   | 20/30 [00:05<00:02,  3.40it/s]
 70%|███████   | 21/30 [00:06<00:02,  3.45it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.50it/s]
 77%|███████▋  | 23/30 [00:06<00:02,  3.41it/s]
 80%|████████  | 24/30 [00:07<00:01,  3.45it/s]
 83%|████████▎ | 25/30 [00:07<00:01,  3.42it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.39it/s]
 90%|█████████ | 27/30 [00:07<00:00,  3.37it/s]
 93%|█████████▎| 28/30 [00:08<00:00,  3.35it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.33it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.51it/s]
 20%|██        | 2/10 [00:00<00:02,  3.51it/s]
 30%|███       | 3/10 [00:00<00:02,  3.33it/s]
 40%|████      | 4/10 [00:01<00:01,  3.36it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.40it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.33it/s]
 70%|███████   | 7/10 [00:02<00:00,  3.37it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.39it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.42it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.57it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.51it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.41it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.28it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.27it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.40it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.55it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.49it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.45it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.46it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.41it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.47it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:08,  3.74it/s]
  6%|▋         | 2/32 [00:00<00:08,  3.59it/s]
  9%|▉         | 3/32 [00:00<00:08,  3.59it/s]
 12%|█▎        | 4/32 [00:01<00:07,  3.60it/s]
 16%|█▌        | 5/32 [00:01<00:07,  3.63it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.55it/s]
 22%|██▏       | 7/32 [00:01<00:06,  3.58it/s]
 25%|██▌       | 8/32 [00:02<00:06,  3.54it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.57it/s]
 31%|███▏      | 10/32 [00:02<00:06,  3.59it/s]
 34%|███▍      | 11/32 [00:03<00:05,  3.60it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.59it/s]
 41%|████      | 13/32 [00:03<00:05,  3.55it/s]
 44%|████▍     | 14/32 [00:03<00:05,  3.52it/s]
 47%|████▋     | 15/32 [00:04<00:04,  3.53it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.52it/s]
 53%|█████▎    | 17/32 [00:04<00:04,  3.54it/s]
 56%|█████▋    | 18/32 [00:05<00:03,  3.54it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.55it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.53it/s]
 66%|██████▌   | 21/32 [00:05<00:03,  3.52it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.60it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.61it/s]
 75%|███████▌  | 24/32 [00:06<00:02,  3.57it/s]
 78%|███████▊  | 25/32 [00:07<00:01,  3.52it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.56it/s]
 84%|████████▍ | 27/32 [00:07<00:01,  3.54it/s]
 88%|████████▊ | 28/32 [00:07<00:01,  3.57it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.58it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.58it/s]
 97%|█████████▋| 31/32 [00:08<00:00,  3.52it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:05,  3.43it/s]
 11%|█         | 2/19 [00:00<00:04,  3.55it/s]
 16%|█▌        | 3/19 [00:00<00:04,  3.53it/s]
 21%|██        | 4/19 [00:01<00:04,  3.51it/s]
 26%|██▋       | 5/19 [00:01<00:04,  3.11it/s]
 32%|███▏      | 6/19 [00:01<00:04,  3.20it/s]
 37%|███▋      | 7/19 [00:02<00:03,  3.34it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.35it/s]
 47%|████▋     | 9/19 [00:02<00:02,  3.41it/s]
 53%|█████▎    | 10/19 [00:02<00:02,  3.41it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.44it/s]
 63%|██████▎   | 12/19 [00:03<00:02,  3.48it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.42it/s]
 74%|███████▎  | 14/19 [00:04<00:01,  3.41it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.48it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.51it/s]
 89%|████████▉ | 17/19 [00:04<00:00,  3.52it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.44it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.38it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.45it/s]
 11%|█         | 3/27 [00:00<00:06,  3.52it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.46it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.48it/s]
 22%|██▏       | 6/27 [00:01<00:06,  3.49it/s]
 26%|██▌       | 7/27 [00:02<00:06,  3.31it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.33it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.40it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.42it/s]
 41%|████      | 11/27 [00:03<00:04,  3.42it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.40it/s]
 48%|████▊     | 13/27 [00:03<00:04,  3.46it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.45it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.51it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.54it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.48it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.44it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.44it/s]
 74%|███████▍  | 20/27 [00:05<00:02,  3.49it/s]
 78%|███████▊  | 21/27 [00:06<00:01,  3.49it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.51it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.53it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.52it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.54it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.56it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.45it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.44it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.47it/s]
 31%|███       | 4/13 [00:01<00:02,  3.51it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.51it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.52it/s]
 54%|█████▍    | 7/13 [00:02<00:01,  3.50it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.44it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.42it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.39it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.38it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.43it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.33it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.11it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.42it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.41it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.44it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.41it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.44it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.45it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.48it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.38it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.42it/s]
 12%|█▎        | 3/24 [00:00<00:06,  3.41it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.40it/s]
 21%|██        | 5/24 [00:01<00:05,  3.43it/s]
 25%|██▌       | 6/24 [00:01<00:05,  3.44it/s]
 29%|██▉       | 7/24 [00:02<00:04,  3.44it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.45it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.44it/s]
 42%|████▏     | 10/24 [00:02<00:04,  3.41it/s]
 46%|████▌     | 11/24 [00:03<00:03,  3.42it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.42it/s]
 54%|█████▍    | 13/24 [00:03<00:03,  3.44it/s]
 58%|█████▊    | 14/24 [00:04<00:02,  3.42it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.45it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.47it/s]
 71%|███████   | 17/24 [00:04<00:02,  3.46it/s]
 75%|███████▌  | 18/24 [00:05<00:01,  3.47it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.45it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.51it/s]
 88%|████████▊ | 21/24 [00:06<00:00,  3.49it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.45it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.44it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.46it/s]
 22%|██▏       | 2/9 [00:00<00:02,  3.45it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.39it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.37it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.42it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.47it/s]
 78%|███████▊  | 7/9 [00:02<00:00,  3.54it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.53it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.35it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.45it/s]
 11%|█         | 3/27 [00:00<00:06,  3.45it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.48it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.47it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.55it/s]
 26%|██▌       | 7/27 [00:02<00:05,  3.49it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.49it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.52it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.51it/s]
 41%|████      | 11/27 [00:03<00:04,  3.47it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.46it/s]
 48%|████▊     | 13/27 [00:03<00:04,  3.47it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.48it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.47it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.47it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.48it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.49it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.47it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.50it/s]
 78%|███████▊  | 21/27 [00:06<00:01,  3.48it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.44it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.46it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.46it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.42it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.44it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  3.51it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.38it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.37it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.33it/s]
 42%|████▏     | 5/12 [00:01<00:02,  3.43it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.45it/s]
 58%|█████▊    | 7/12 [00:02<00:01,  3.48it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.49it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.53it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.51it/s]
 92%|█████████▏| 11/12 [00:03<00:00,  3.54it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.58it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.57it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.53it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.55it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.57it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.53it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.50it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.48it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.46it/s]
 31%|███       | 4/13 [00:01<00:02,  3.44it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.47it/s]
 46%|████▌     | 6/13 [00:01<00:02,  3.46it/s]
 54%|█████▍    | 7/13 [00:02<00:01,  3.43it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.47it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.49it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.47it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.52it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.46it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.51it/s]
 10%|█         | 2/20 [00:00<00:05,  3.49it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.48it/s]
 20%|██        | 4/20 [00:01<00:04,  3.50it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.57it/s]
 30%|███       | 6/20 [00:01<00:03,  3.54it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.51it/s]
 40%|████      | 8/20 [00:02<00:03,  3.56it/s]
 45%|████▌     | 9/20 [00:02<00:03,  3.54it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.50it/s]
 55%|█████▌    | 11/20 [00:03<00:02,  3.57it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.55it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.62it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.63it/s]
 75%|███████▌  | 15/20 [00:04<00:01,  3.57it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.61it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.56it/s]
 90%|█████████ | 18/20 [00:05<00:00,  3.60it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.56it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:04,  3.33it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.46it/s]
 20%|██        | 3/15 [00:00<00:03,  3.48it/s]
 27%|██▋       | 4/15 [00:01<00:03,  3.55it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.51it/s]
 40%|████      | 6/15 [00:01<00:02,  3.61it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.54it/s]
 53%|█████▎    | 8/15 [00:02<00:02,  3.50it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.49it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.47it/s]
 73%|███████▎  | 11/15 [00:03<00:01,  3.47it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.51it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.55it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.58it/s]
+
+[2025-07-11 17:11:15] Weighted Per image PCK0.10: 79.62%, image PCK0.05: 64.95%, image PCK0.01: 86.19%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.5_layer_12.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:11:21] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.5_layer_12', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.5_layer_12'], NO_DINO=False, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='6', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.5_layer_12']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:20,  1.45s/it]
 13%|█▎        | 2/15 [00:01<00:11,  1.09it/s]
 20%|██        | 3/15 [00:02<00:08,  1.48it/s]
 27%|██▋       | 4/15 [00:02<00:05,  1.88it/s]
 33%|███▎      | 5/15 [00:02<00:04,  2.25it/s]
 40%|████      | 6/15 [00:03<00:03,  2.52it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.80it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  3.01it/s]
 60%|██████    | 9/15 [00:04<00:01,  3.13it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.21it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.30it/s]
 80%|████████  | 12/15 [00:04<00:00,  3.34it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.46it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.49it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:09,  3.19it/s]
  7%|▋         | 2/30 [00:00<00:08,  3.20it/s]
 10%|█         | 3/30 [00:00<00:08,  3.31it/s]
 13%|█▎        | 4/30 [00:01<00:07,  3.36it/s]
 17%|█▋        | 5/30 [00:01<00:07,  3.46it/s]
 20%|██        | 6/30 [00:01<00:06,  3.49it/s]
 23%|██▎       | 7/30 [00:02<00:06,  3.48it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.34it/s]
 30%|███       | 9/30 [00:02<00:06,  3.40it/s]
 33%|███▎      | 10/30 [00:02<00:05,  3.40it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.46it/s]
 40%|████      | 12/30 [00:03<00:05,  3.45it/s]
 43%|████▎     | 13/30 [00:03<00:04,  3.49it/s]
 47%|████▋     | 14/30 [00:04<00:04,  3.46it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.50it/s]
 53%|█████▎    | 16/30 [00:04<00:03,  3.55it/s]
 57%|█████▋    | 17/30 [00:04<00:03,  3.62it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.61it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.58it/s]
 67%|██████▋   | 20/30 [00:05<00:02,  3.56it/s]
 70%|███████   | 21/30 [00:06<00:02,  3.49it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.45it/s]
 77%|███████▋  | 23/30 [00:06<00:01,  3.54it/s]
 80%|████████  | 24/30 [00:06<00:01,  3.50it/s]
 83%|████████▎ | 25/30 [00:07<00:01,  3.50it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.54it/s]
 90%|█████████ | 27/30 [00:07<00:00,  3.53it/s]
 93%|█████████▎| 28/30 [00:08<00:00,  3.55it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.57it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.59it/s]
 20%|██        | 2/10 [00:00<00:02,  3.53it/s]
 30%|███       | 3/10 [00:00<00:02,  3.30it/s]
 40%|████      | 4/10 [00:01<00:01,  3.30it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.32it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.42it/s]
 70%|███████   | 7/10 [00:02<00:00,  3.45it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.41it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.39it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.25it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.32it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.41it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.49it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.56it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.49it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.47it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.52it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.53it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.53it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.56it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.51it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:09,  3.35it/s]
  6%|▋         | 2/32 [00:00<00:08,  3.42it/s]
  9%|▉         | 3/32 [00:00<00:08,  3.39it/s]
 12%|█▎        | 4/32 [00:01<00:08,  3.45it/s]
 16%|█▌        | 5/32 [00:01<00:07,  3.43it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.49it/s]
 22%|██▏       | 7/32 [00:02<00:07,  3.48it/s]
 25%|██▌       | 8/32 [00:02<00:06,  3.55it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.55it/s]
 31%|███▏      | 10/32 [00:02<00:06,  3.49it/s]
 34%|███▍      | 11/32 [00:03<00:06,  3.47it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.45it/s]
 41%|████      | 13/32 [00:03<00:05,  3.50it/s]
 44%|████▍     | 14/32 [00:04<00:05,  3.53it/s]
 47%|████▋     | 15/32 [00:04<00:04,  3.52it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.51it/s]
 53%|█████▎    | 17/32 [00:04<00:04,  3.48it/s]
 56%|█████▋    | 18/32 [00:05<00:04,  3.49it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.49it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.49it/s]
 66%|██████▌   | 21/32 [00:06<00:03,  3.50it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.46it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.47it/s]
 75%|███████▌  | 24/32 [00:06<00:02,  3.51it/s]
 78%|███████▊  | 25/32 [00:07<00:01,  3.55it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.50it/s]
 84%|████████▍ | 27/32 [00:07<00:01,  3.51it/s]
 88%|████████▊ | 28/32 [00:08<00:01,  3.45it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.44it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.43it/s]
 97%|█████████▋| 31/32 [00:08<00:00,  3.48it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:04,  3.64it/s]
 11%|█         | 2/19 [00:00<00:04,  3.60it/s]
 16%|█▌        | 3/19 [00:00<00:04,  3.56it/s]
 21%|██        | 4/19 [00:01<00:04,  3.51it/s]
 26%|██▋       | 5/19 [00:01<00:03,  3.52it/s]
 32%|███▏      | 6/19 [00:01<00:03,  3.45it/s]
 37%|███▋      | 7/19 [00:02<00:03,  3.39it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.42it/s]
 47%|████▋     | 9/19 [00:02<00:02,  3.39it/s]
 53%|█████▎    | 10/19 [00:02<00:02,  3.45it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.46it/s]
 63%|██████▎   | 12/19 [00:03<00:02,  3.42it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.43it/s]
 74%|███████▎  | 14/19 [00:04<00:01,  3.46it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.43it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.39it/s]
 89%|████████▉ | 17/19 [00:04<00:00,  3.37it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.41it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.57it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.44it/s]
 11%|█         | 3/27 [00:00<00:07,  3.41it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.42it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.37it/s]
 22%|██▏       | 6/27 [00:01<00:06,  3.39it/s]
 26%|██▌       | 7/27 [00:02<00:05,  3.41it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.39it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.37it/s]
 37%|███▋      | 10/27 [00:02<00:05,  3.39it/s]
 41%|████      | 11/27 [00:03<00:04,  3.41it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.41it/s]
 48%|████▊     | 13/27 [00:03<00:04,  3.38it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.39it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.34it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.32it/s]
 63%|██████▎   | 17/27 [00:05<00:02,  3.34it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.44it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.45it/s]
 74%|███████▍  | 20/27 [00:05<00:02,  3.41it/s]
 78%|███████▊  | 21/27 [00:06<00:01,  3.43it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.44it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.46it/s]
 89%|████████▉ | 24/27 [00:07<00:00,  3.42it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.41it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.39it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.29it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.51it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.48it/s]
 31%|███       | 4/13 [00:01<00:02,  3.39it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.37it/s]
 46%|████▌     | 6/13 [00:01<00:02,  3.33it/s]
 54%|█████▍    | 7/13 [00:02<00:01,  3.41it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.43it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.46it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.52it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.53it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.48it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.49it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.54it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.44it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.46it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.47it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.47it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.44it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.47it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.45it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.50it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.49it/s]
 12%|█▎        | 3/24 [00:00<00:06,  3.49it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.56it/s]
 21%|██        | 5/24 [00:01<00:05,  3.51it/s]
 25%|██▌       | 6/24 [00:01<00:05,  3.48it/s]
 29%|██▉       | 7/24 [00:01<00:04,  3.50it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.47it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.47it/s]
 42%|████▏     | 10/24 [00:02<00:04,  3.46it/s]
 46%|████▌     | 11/24 [00:03<00:03,  3.50it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.50it/s]
 54%|█████▍    | 13/24 [00:03<00:03,  3.45it/s]
 58%|█████▊    | 14/24 [00:04<00:02,  3.43it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.40it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.39it/s]
 71%|███████   | 17/24 [00:04<00:02,  3.45it/s]
 75%|███████▌  | 18/24 [00:05<00:01,  3.46it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.49it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.44it/s]
 88%|████████▊ | 21/24 [00:06<00:00,  3.46it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.45it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.45it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.41it/s]
 22%|██▏       | 2/9 [00:00<00:02,  3.45it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.56it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.60it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.54it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.54it/s]
 78%|███████▊  | 7/9 [00:01<00:00,  3.51it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.53it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.50it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.43it/s]
 11%|█         | 3/27 [00:00<00:07,  3.40it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.42it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.43it/s]
 22%|██▏       | 6/27 [00:01<00:06,  3.39it/s]
 26%|██▌       | 7/27 [00:02<00:05,  3.44it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.44it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.44it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.43it/s]
 41%|████      | 11/27 [00:03<00:04,  3.41it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.43it/s]
 48%|████▊     | 13/27 [00:03<00:04,  3.50it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.50it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.47it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.46it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.46it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.46it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.43it/s]
 74%|███████▍  | 20/27 [00:05<00:02,  3.42it/s]
 78%|███████▊  | 21/27 [00:06<00:01,  3.44it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.49it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.45it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.54it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.57it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.49it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  3.53it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.69it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.73it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.78it/s]
 42%|████▏     | 5/12 [00:01<00:01,  3.54it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.58it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.51it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.50it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.50it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.52it/s]
 92%|█████████▏| 11/12 [00:03<00:00,  3.48it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.27it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.36it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.41it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.34it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.28it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.34it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.54it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.58it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.48it/s]
 31%|███       | 4/13 [00:01<00:02,  3.54it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.48it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.51it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.49it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.43it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.39it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.44it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.40it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.41it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.41it/s]
 10%|█         | 2/20 [00:00<00:05,  3.51it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.57it/s]
 20%|██        | 4/20 [00:01<00:04,  3.58it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.55it/s]
 30%|███       | 6/20 [00:01<00:03,  3.59it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.65it/s]
 40%|████      | 8/20 [00:02<00:03,  3.58it/s]
 45%|████▌     | 9/20 [00:02<00:03,  3.51it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.55it/s]
 55%|█████▌    | 11/20 [00:03<00:02,  3.55it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.56it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.52it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.51it/s]
 75%|███████▌  | 15/20 [00:04<00:01,  3.56it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.53it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.61it/s]
 90%|█████████ | 18/20 [00:05<00:00,  3.57it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.59it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:04,  3.37it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.44it/s]
 20%|██        | 3/15 [00:00<00:03,  3.49it/s]
 27%|██▋       | 4/15 [00:01<00:03,  3.50it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.51it/s]
 40%|████      | 6/15 [00:01<00:02,  3.31it/s]
 47%|████▋     | 7/15 [00:02<00:02,  3.34it/s]
 53%|█████▎    | 8/15 [00:02<00:02,  3.41it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.39it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.47it/s]
 73%|███████▎  | 11/15 [00:03<00:01,  3.52it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.55it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.55it/s]
 93%|█████████▎| 14/15 [00:04<00:00,  3.50it/s]
+
+[2025-07-11 17:12:50] Weighted Per image PCK0.10: 79.24%, image PCK0.05: 63.05%, image PCK0.01: 86.40%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.5_layer_12_no_dino.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:11:22] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.5_layer_12_no_dino', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.5_layer_12'], NO_DINO=True, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='7', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.5_layer_12']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:19,  1.37s/it]
 13%|█▎        | 2/15 [00:01<00:11,  1.14it/s]
 20%|██        | 3/15 [00:02<00:07,  1.63it/s]
 27%|██▋       | 4/15 [00:02<00:05,  2.06it/s]
 33%|███▎      | 5/15 [00:02<00:04,  2.41it/s]
 40%|████      | 6/15 [00:03<00:03,  2.69it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.90it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  3.08it/s]
 60%|██████    | 9/15 [00:03<00:01,  3.18it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.29it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.31it/s]
 80%|████████  | 12/15 [00:04<00:00,  3.31it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.28it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.30it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:10,  2.74it/s]
  7%|▋         | 2/30 [00:00<00:09,  2.94it/s]
 10%|█         | 3/30 [00:00<00:08,  3.18it/s]
 13%|█▎        | 4/30 [00:01<00:08,  3.22it/s]
 17%|█▋        | 5/30 [00:01<00:07,  3.20it/s]
 20%|██        | 6/30 [00:01<00:07,  3.22it/s]
 23%|██▎       | 7/30 [00:02<00:07,  3.04it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.20it/s]
 30%|███       | 9/30 [00:02<00:06,  3.26it/s]
 33%|███▎      | 10/30 [00:03<00:06,  3.30it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.30it/s]
 40%|████      | 12/30 [00:03<00:05,  3.34it/s]
 43%|████▎     | 13/30 [00:04<00:05,  3.38it/s]
 47%|████▋     | 14/30 [00:04<00:04,  3.50it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.51it/s]
 53%|█████▎    | 16/30 [00:04<00:03,  3.51it/s]
 57%|█████▋    | 17/30 [00:05<00:03,  3.41it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.43it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.37it/s]
 67%|██████▋   | 20/30 [00:06<00:02,  3.37it/s]
 70%|███████   | 21/30 [00:06<00:02,  3.45it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.44it/s]
 77%|███████▋  | 23/30 [00:06<00:02,  3.41it/s]
 80%|████████  | 24/30 [00:07<00:01,  3.43it/s]
 83%|████████▎ | 25/30 [00:07<00:01,  3.37it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.33it/s]
 90%|█████████ | 27/30 [00:08<00:00,  3.34it/s]
 93%|█████████▎| 28/30 [00:08<00:00,  3.33it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.34it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.41it/s]
 20%|██        | 2/10 [00:00<00:02,  3.48it/s]
 30%|███       | 3/10 [00:00<00:02,  3.41it/s]
 40%|████      | 4/10 [00:01<00:01,  3.44it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.47it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.41it/s]
 70%|███████   | 7/10 [00:02<00:00,  3.41it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.33it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.38it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.43it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.47it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.39it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.28it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.31it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.45it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.60it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.54it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.50it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.41it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.40it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.41it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:11,  2.64it/s]
  6%|▋         | 2/32 [00:00<00:10,  2.78it/s]
  9%|▉         | 3/32 [00:01<00:09,  3.02it/s]
 12%|█▎        | 4/32 [00:01<00:08,  3.16it/s]
 16%|█▌        | 5/32 [00:01<00:08,  3.15it/s]
 19%|█▉        | 6/32 [00:01<00:08,  3.18it/s]
 22%|██▏       | 7/32 [00:02<00:07,  3.28it/s]
 25%|██▌       | 8/32 [00:02<00:07,  3.29it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.34it/s]
 31%|███▏      | 10/32 [00:03<00:06,  3.23it/s]
 34%|███▍      | 11/32 [00:03<00:06,  3.33it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.38it/s]
 41%|████      | 13/32 [00:04<00:05,  3.31it/s]
 44%|████▍     | 14/32 [00:04<00:05,  3.32it/s]
 47%|████▋     | 15/32 [00:04<00:05,  3.31it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.35it/s]
 53%|█████▎    | 17/32 [00:05<00:04,  3.45it/s]
 56%|█████▋    | 18/32 [00:05<00:04,  3.37it/s]
 59%|█████▉    | 19/32 [00:05<00:04,  3.25it/s]
 62%|██████▎   | 20/32 [00:06<00:03,  3.10it/s]
 66%|██████▌   | 21/32 [00:06<00:03,  3.14it/s]
 69%|██████▉   | 22/32 [00:06<00:03,  3.30it/s]
 72%|███████▏  | 23/32 [00:07<00:02,  3.41it/s]
 75%|███████▌  | 24/32 [00:07<00:02,  3.44it/s]
 78%|███████▊  | 25/32 [00:07<00:02,  3.44it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.52it/s]
 84%|████████▍ | 27/32 [00:08<00:01,  3.39it/s]
 88%|████████▊ | 28/32 [00:08<00:01,  3.42it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.49it/s]
 94%|█████████▍| 30/32 [00:09<00:00,  3.52it/s]
 97%|█████████▋| 31/32 [00:09<00:00,  3.46it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:05,  3.20it/s]
 11%|█         | 2/19 [00:00<00:04,  3.43it/s]
 16%|█▌        | 3/19 [00:00<00:04,  3.49it/s]
 21%|██        | 4/19 [00:01<00:04,  3.17it/s]
 26%|██▋       | 5/19 [00:01<00:04,  3.27it/s]
 32%|███▏      | 6/19 [00:01<00:03,  3.40it/s]
 37%|███▋      | 7/19 [00:02<00:03,  3.55it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.47it/s]
 47%|████▋     | 9/19 [00:02<00:02,  3.51it/s]
 53%|█████▎    | 10/19 [00:02<00:02,  3.50it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.53it/s]
 63%|██████▎   | 12/19 [00:03<00:01,  3.53it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.48it/s]
 74%|███████▎  | 14/19 [00:04<00:01,  3.50it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.57it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.58it/s]
 89%|████████▉ | 17/19 [00:04<00:00,  3.59it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.53it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.45it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.46it/s]
 11%|█         | 3/27 [00:00<00:06,  3.59it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.51it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.30it/s]
 22%|██▏       | 6/27 [00:01<00:06,  3.27it/s]
 26%|██▌       | 7/27 [00:02<00:05,  3.35it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.26it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.33it/s]
 37%|███▋      | 10/27 [00:02<00:05,  3.32it/s]
 41%|████      | 11/27 [00:03<00:04,  3.33it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.36it/s]
 48%|████▊     | 13/27 [00:03<00:04,  3.42it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.40it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.43it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.51it/s]
 63%|██████▎   | 17/27 [00:05<00:02,  3.40it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.37it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.39it/s]
 74%|███████▍  | 20/27 [00:05<00:02,  3.49it/s]
 78%|███████▊  | 21/27 [00:06<00:01,  3.53it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.59it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.47it/s]
 89%|████████▉ | 24/27 [00:07<00:00,  3.53it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.43it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.43it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:04,  2.89it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.12it/s]
 23%|██▎       | 3/13 [00:00<00:03,  3.20it/s]
 31%|███       | 4/13 [00:01<00:02,  3.17it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.24it/s]
 46%|████▌     | 6/13 [00:01<00:02,  3.30it/s]
 54%|█████▍    | 7/13 [00:02<00:01,  3.33it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.33it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.33it/s]
 77%|███████▋  | 10/13 [00:03<00:00,  3.35it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.38it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.47it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.15it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.30it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.34it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.35it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.41it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.39it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.42it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.43it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.43it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.47it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.55it/s]
 12%|█▎        | 3/24 [00:00<00:05,  3.60it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.54it/s]
 21%|██        | 5/24 [00:01<00:05,  3.56it/s]
 25%|██▌       | 6/24 [00:01<00:05,  3.59it/s]
 29%|██▉       | 7/24 [00:01<00:04,  3.61it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.61it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.59it/s]
 42%|████▏     | 10/24 [00:02<00:03,  3.57it/s]
 46%|████▌     | 11/24 [00:03<00:03,  3.57it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.59it/s]
 54%|█████▍    | 13/24 [00:03<00:03,  3.62it/s]
 58%|█████▊    | 14/24 [00:03<00:02,  3.62it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.67it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.64it/s]
 71%|███████   | 17/24 [00:04<00:01,  3.61it/s]
 75%|███████▌  | 18/24 [00:04<00:01,  3.64it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.61it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.67it/s]
 88%|████████▊ | 21/24 [00:05<00:00,  3.62it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.61it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.58it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.39it/s]
 22%|██▏       | 2/9 [00:00<00:02,  3.50it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.51it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.54it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.62it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.65it/s]
 78%|███████▊  | 7/9 [00:01<00:00,  3.74it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.70it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.50it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.59it/s]
 11%|█         | 3/27 [00:00<00:06,  3.61it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.60it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.60it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.67it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.61it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.61it/s]
 33%|███▎      | 9/27 [00:02<00:04,  3.64it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.65it/s]
 41%|████      | 11/27 [00:03<00:04,  3.62it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.59it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.56it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.55it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.56it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.57it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.60it/s]
 67%|██████▋   | 18/27 [00:04<00:02,  3.64it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.64it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.65it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.63it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.59it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.59it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.60it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.56it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.58it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  3.54it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.54it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.56it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.55it/s]
 42%|████▏     | 5/12 [00:01<00:01,  3.66it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.68it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.75it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.72it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.75it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.73it/s]
 92%|█████████▏| 11/12 [00:02<00:00,  3.75it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.69it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.79it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.76it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.77it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.78it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.73it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.66it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.66it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.66it/s]
 31%|███       | 4/13 [00:01<00:02,  3.60it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.61it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.61it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.60it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.62it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.65it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.60it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.65it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.59it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.64it/s]
 10%|█         | 2/20 [00:00<00:04,  3.71it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.70it/s]
 20%|██        | 4/20 [00:01<00:04,  3.70it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.74it/s]
 30%|███       | 6/20 [00:01<00:03,  3.73it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.71it/s]
 40%|████      | 8/20 [00:02<00:03,  3.79it/s]
 45%|████▌     | 9/20 [00:02<00:02,  3.84it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.78it/s]
 55%|█████▌    | 11/20 [00:02<00:02,  3.83it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.81it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.85it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.87it/s]
 75%|███████▌  | 15/20 [00:03<00:01,  3.84it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.85it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.74it/s]
 90%|█████████ | 18/20 [00:04<00:00,  3.75it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.69it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:03,  3.56it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.76it/s]
 20%|██        | 3/15 [00:00<00:03,  3.73it/s]
 27%|██▋       | 4/15 [00:01<00:02,  3.80it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.78it/s]
 40%|████      | 6/15 [00:01<00:02,  3.85it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.79it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.75it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.73it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.73it/s]
 73%|███████▎  | 11/15 [00:02<00:01,  3.74it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.76it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.78it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.80it/s]
+
+[2025-07-11 17:12:50] Weighted Per image PCK0.10: 73.39%, image PCK0.05: 57.35%, image PCK0.01: 82.38%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.5_layer_13_no_dino.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:12:52] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.5_layer_13_no_dino', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.5_layer_13'], NO_DINO=True, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='5', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.5_layer_13']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:20,  1.49s/it]
 13%|█▎        | 2/15 [00:02<00:12,  1.04it/s]
 20%|██        | 3/15 [00:02<00:08,  1.38it/s]
 27%|██▋       | 4/15 [00:02<00:06,  1.81it/s]
 33%|███▎      | 5/15 [00:03<00:04,  2.16it/s]
 40%|████      | 6/15 [00:03<00:03,  2.50it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.77it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  2.97it/s]
 60%|██████    | 9/15 [00:04<00:01,  3.14it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.29it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  2.83it/s]
 80%|████████  | 12/15 [00:05<00:00,  3.01it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.10it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.29it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:16,  1.79it/s]
  7%|▋         | 2/30 [00:00<00:11,  2.43it/s]
 10%|█         | 3/30 [00:01<00:09,  2.83it/s]
 13%|█▎        | 4/30 [00:01<00:08,  3.09it/s]
 17%|█▋        | 5/30 [00:01<00:07,  3.14it/s]
 20%|██        | 6/30 [00:02<00:08,  2.79it/s]
 23%|██▎       | 7/30 [00:02<00:07,  2.97it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.17it/s]
 30%|███       | 9/30 [00:03<00:06,  3.26it/s]
 33%|███▎      | 10/30 [00:03<00:06,  3.00it/s]
 37%|███▋      | 11/30 [00:03<00:06,  3.08it/s]
 40%|████      | 12/30 [00:04<00:06,  2.75it/s]
 43%|████▎     | 13/30 [00:04<00:05,  2.97it/s]
 47%|████▋     | 14/30 [00:04<00:05,  3.14it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.30it/s]
 53%|█████▎    | 16/30 [00:05<00:05,  2.57it/s]
 57%|█████▋    | 17/30 [00:05<00:04,  2.78it/s]
 60%|██████    | 18/30 [00:06<00:04,  2.96it/s]
 63%|██████▎   | 19/30 [00:06<00:03,  3.05it/s]
 67%|██████▋   | 20/30 [00:06<00:03,  3.10it/s]
 70%|███████   | 21/30 [00:07<00:03,  2.94it/s]
 73%|███████▎  | 22/30 [00:07<00:02,  3.11it/s]
 77%|███████▋  | 23/30 [00:07<00:02,  3.22it/s]
 80%|████████  | 24/30 [00:07<00:01,  3.37it/s]
 83%|████████▎ | 25/30 [00:08<00:01,  3.44it/s]
 87%|████████▋ | 26/30 [00:08<00:01,  3.11it/s]
 90%|█████████ | 27/30 [00:08<00:00,  3.20it/s]
 93%|█████████▎| 28/30 [00:09<00:00,  3.26it/s]
 97%|█████████▋| 29/30 [00:09<00:00,  3.32it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:03,  2.66it/s]
 20%|██        | 2/10 [00:00<00:02,  2.85it/s]
 30%|███       | 3/10 [00:01<00:02,  3.03it/s]
 40%|████      | 4/10 [00:01<00:01,  3.16it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.27it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.32it/s]
 70%|███████   | 7/10 [00:02<00:00,  3.37it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.41it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.47it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.26it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.39it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.38it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.46it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  2.79it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  2.80it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.23it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.27it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.10it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.26it/s]
 75%|███████▌  | 6/8 [00:02<00:00,  2.83it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.03it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:12,  2.55it/s]
  6%|▋         | 2/32 [00:00<00:09,  3.11it/s]
  9%|▉         | 3/32 [00:00<00:08,  3.25it/s]
 12%|█▎        | 4/32 [00:01<00:08,  3.35it/s]
 16%|█▌        | 5/32 [00:01<00:08,  3.19it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.30it/s]
 22%|██▏       | 7/32 [00:02<00:08,  3.10it/s]
 25%|██▌       | 8/32 [00:02<00:07,  3.15it/s]
 28%|██▊       | 9/32 [00:02<00:07,  3.25it/s]
 31%|███▏      | 10/32 [00:03<00:07,  2.82it/s]
 34%|███▍      | 11/32 [00:03<00:07,  2.75it/s]
 38%|███▊      | 12/32 [00:03<00:06,  3.01it/s]
 41%|████      | 13/32 [00:04<00:06,  2.79it/s]
 44%|████▍     | 14/32 [00:04<00:05,  3.00it/s]
 47%|████▋     | 15/32 [00:04<00:05,  3.21it/s]
 50%|█████     | 16/32 [00:05<00:04,  3.37it/s]
 53%|█████▎    | 17/32 [00:05<00:04,  3.46it/s]
 56%|█████▋    | 18/32 [00:05<00:03,  3.57it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.60it/s]
 62%|██████▎   | 20/32 [00:06<00:03,  3.28it/s]
 66%|██████▌   | 21/32 [00:06<00:03,  3.34it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.50it/s]
 72%|███████▏  | 23/32 [00:07<00:02,  3.57it/s]
 75%|███████▌  | 24/32 [00:07<00:02,  3.57it/s]
 78%|███████▊  | 25/32 [00:07<00:02,  3.37it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.51it/s]
 84%|████████▍ | 27/32 [00:08<00:01,  3.54it/s]
 88%|████████▊ | 28/32 [00:08<00:01,  3.62it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.67it/s]
 94%|█████████▍| 30/32 [00:09<00:00,  3.69it/s]
 97%|█████████▋| 31/32 [00:09<00:00,  3.64it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:07,  2.44it/s]
 11%|█         | 2/19 [00:00<00:06,  2.56it/s]
 16%|█▌        | 3/19 [00:01<00:06,  2.56it/s]
 21%|██        | 4/19 [00:01<00:05,  2.71it/s]
 26%|██▋       | 5/19 [00:01<00:04,  2.95it/s]
 32%|███▏      | 6/19 [00:02<00:04,  2.72it/s]
 37%|███▋      | 7/19 [00:02<00:03,  3.05it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.13it/s]
 47%|████▋     | 9/19 [00:03<00:03,  2.95it/s]
 53%|█████▎    | 10/19 [00:03<00:02,  3.09it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  2.84it/s]
 63%|██████▎   | 12/19 [00:04<00:02,  2.73it/s]
 68%|██████▊   | 13/19 [00:04<00:02,  2.86it/s]
 74%|███████▎  | 14/19 [00:04<00:01,  3.05it/s]
 79%|███████▉  | 15/19 [00:05<00:01,  3.18it/s]
 84%|████████▍ | 16/19 [00:05<00:00,  3.33it/s]
 89%|████████▉ | 17/19 [00:05<00:00,  3.34it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.33it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:09,  2.71it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.21it/s]
 11%|█         | 3/27 [00:00<00:06,  3.43it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.41it/s]
 19%|█▊        | 5/27 [00:01<00:07,  3.10it/s]
 22%|██▏       | 6/27 [00:01<00:06,  3.23it/s]
 26%|██▌       | 7/27 [00:02<00:06,  3.18it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.24it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.25it/s]
 37%|███▋      | 10/27 [00:03<00:05,  3.33it/s]
 41%|████      | 11/27 [00:03<00:04,  3.38it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.33it/s]
 48%|████▊     | 13/27 [00:03<00:04,  3.37it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.44it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.53it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.45it/s]
 63%|██████▎   | 17/27 [00:05<00:02,  3.46it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.49it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.39it/s]
 74%|███████▍  | 20/27 [00:05<00:02,  3.45it/s]
 78%|███████▊  | 21/27 [00:06<00:01,  3.54it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.48it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.22it/s]
 89%|████████▉ | 24/27 [00:07<00:01,  2.97it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.16it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.07it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.64it/s]
 15%|█▌        | 2/13 [00:00<00:03,  2.90it/s]
 23%|██▎       | 3/13 [00:00<00:03,  3.21it/s]
 31%|███       | 4/13 [00:01<00:03,  2.56it/s]
 38%|███▊      | 5/13 [00:01<00:02,  2.92it/s]
 46%|████▌     | 6/13 [00:02<00:03,  2.28it/s]
 54%|█████▍    | 7/13 [00:02<00:02,  2.56it/s]
 62%|██████▏   | 8/13 [00:03<00:01,  2.52it/s]
 69%|██████▉   | 9/13 [00:03<00:01,  2.61it/s]
 77%|███████▋  | 10/13 [00:03<00:01,  2.73it/s]
 85%|████████▍ | 11/13 [00:04<00:00,  2.90it/s]
 92%|█████████▏| 12/13 [00:04<00:00,  2.72it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.51it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.33it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  2.60it/s]
 25%|██▌       | 2/8 [00:00<00:02,  2.97it/s]
 38%|███▊      | 3/8 [00:01<00:01,  3.05it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.17it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.30it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.39it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.51it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:07,  2.99it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.24it/s]
 12%|█▎        | 3/24 [00:00<00:06,  3.37it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.37it/s]
 21%|██        | 5/24 [00:01<00:05,  3.19it/s]
 25%|██▌       | 6/24 [00:01<00:05,  3.16it/s]
 29%|██▉       | 7/24 [00:02<00:05,  3.23it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.30it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.17it/s]
 42%|████▏     | 10/24 [00:03<00:04,  3.13it/s]
 46%|████▌     | 11/24 [00:03<00:04,  3.21it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.10it/s]
 54%|█████▍    | 13/24 [00:04<00:03,  3.11it/s]
 58%|█████▊    | 14/24 [00:04<00:03,  3.18it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.24it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.25it/s]
 71%|███████   | 17/24 [00:05<00:02,  3.32it/s]
 75%|███████▌  | 18/24 [00:05<00:01,  3.23it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.27it/s]
 83%|████████▎ | 20/24 [00:06<00:01,  3.34it/s]
 88%|████████▊ | 21/24 [00:06<00:00,  3.38it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.38it/s]
 96%|█████████▌| 23/24 [00:07<00:00,  3.40it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.66it/s]
 22%|██▏       | 2/9 [00:00<00:01,  3.68it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.06it/s]
 44%|████▍     | 4/9 [00:01<00:01,  2.73it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  2.88it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.04it/s]
 78%|███████▊  | 7/9 [00:02<00:00,  3.25it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.32it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:13,  1.91it/s]
  7%|▋         | 2/27 [00:00<00:09,  2.66it/s]
 11%|█         | 3/27 [00:01<00:09,  2.53it/s]
 15%|█▍        | 4/27 [00:01<00:09,  2.44it/s]
 19%|█▊        | 5/27 [00:02<00:09,  2.42it/s]
 22%|██▏       | 6/27 [00:02<00:07,  2.65it/s]
 26%|██▌       | 7/27 [00:02<00:07,  2.84it/s]
 30%|██▉       | 8/27 [00:02<00:06,  3.03it/s]
 33%|███▎      | 9/27 [00:03<00:05,  3.20it/s]
 37%|███▋      | 10/27 [00:03<00:05,  3.37it/s]
 41%|████      | 11/27 [00:03<00:04,  3.41it/s]
 44%|████▍     | 12/27 [00:04<00:04,  3.42it/s]
 48%|████▊     | 13/27 [00:04<00:03,  3.51it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.47it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.53it/s]
 59%|█████▉    | 16/27 [00:05<00:03,  3.51it/s]
 63%|██████▎   | 17/27 [00:05<00:02,  3.56it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.57it/s]
 70%|███████   | 19/27 [00:06<00:02,  3.55it/s]
 74%|███████▍  | 20/27 [00:06<00:01,  3.53it/s]
 78%|███████▊  | 21/27 [00:06<00:01,  3.52it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.48it/s]
 85%|████████▌ | 23/27 [00:07<00:01,  3.47it/s]
 89%|████████▉ | 24/27 [00:07<00:00,  3.47it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.43it/s]
 96%|█████████▋| 26/27 [00:08<00:00,  3.46it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  2.81it/s]
 17%|█▋        | 2/12 [00:00<00:03,  3.07it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.24it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.26it/s]
 42%|████▏     | 5/12 [00:01<00:02,  3.36it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.44it/s]
 58%|█████▊    | 7/12 [00:02<00:01,  3.51it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.51it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.58it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.58it/s]
 92%|█████████▏| 11/12 [00:03<00:00,  3.64it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.55it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.67it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.64it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.69it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.74it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.68it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.28it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.38it/s]
 23%|██▎       | 3/13 [00:00<00:03,  3.27it/s]
 31%|███       | 4/13 [00:01<00:02,  3.11it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.13it/s]
 46%|████▌     | 6/13 [00:01<00:02,  3.21it/s]
 54%|█████▍    | 7/13 [00:02<00:01,  3.20it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.17it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.26it/s]
 77%|███████▋  | 10/13 [00:03<00:00,  3.13it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.25it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.28it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.35it/s]
 10%|█         | 2/20 [00:00<00:05,  3.48it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.56it/s]
 20%|██        | 4/20 [00:01<00:04,  3.63it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.72it/s]
 30%|███       | 6/20 [00:01<00:03,  3.73it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.72it/s]
 40%|████      | 8/20 [00:02<00:03,  3.77it/s]
 45%|████▌     | 9/20 [00:02<00:02,  3.81it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.79it/s]
 55%|█████▌    | 11/20 [00:02<00:02,  3.83it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.79it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.85it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.88it/s]
 75%|███████▌  | 15/20 [00:03<00:01,  3.82it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.78it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.67it/s]
 90%|█████████ | 18/20 [00:04<00:00,  3.71it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.69it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:03,  3.62it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.78it/s]
 20%|██        | 3/15 [00:00<00:03,  3.75it/s]
 27%|██▋       | 4/15 [00:01<00:02,  3.80it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.77it/s]
 40%|████      | 6/15 [00:01<00:02,  3.85it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.77it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.71it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.68it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.64it/s]
 73%|███████▎  | 11/15 [00:02<00:01,  3.66it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.67it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.72it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.74it/s]
+
+[2025-07-11 17:14:26] Weighted Per image PCK0.10: 69.81%, image PCK0.05: 53.87%, image PCK0.01: 79.05%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.5_layer_18_no_dino.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:14:28] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.5_layer_18_no_dino', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.5_layer_18'], NO_DINO=True, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='5', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.5_layer_18']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:19,  1.38s/it]
 13%|█▎        | 2/15 [00:01<00:09,  1.38it/s]
 20%|██        | 3/15 [00:02<00:07,  1.56it/s]
 27%|██▋       | 4/15 [00:02<00:05,  1.98it/s]
 33%|███▎      | 5/15 [00:02<00:04,  2.30it/s]
 40%|████      | 6/15 [00:03<00:03,  2.63it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.84it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  3.07it/s]
 60%|██████    | 9/15 [00:03<00:01,  3.27it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.39it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.46it/s]
 80%|████████  | 12/15 [00:04<00:00,  3.47it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.50it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.05it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:09,  3.15it/s]
  7%|▋         | 2/30 [00:00<00:08,  3.22it/s]
 10%|█         | 3/30 [00:00<00:08,  3.32it/s]
 13%|█▎        | 4/30 [00:01<00:07,  3.35it/s]
 17%|█▋        | 5/30 [00:01<00:07,  3.36it/s]
 20%|██        | 6/30 [00:01<00:07,  3.16it/s]
 23%|██▎       | 7/30 [00:02<00:07,  3.22it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.38it/s]
 30%|███       | 9/30 [00:02<00:06,  3.47it/s]
 33%|███▎      | 10/30 [00:02<00:05,  3.58it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.29it/s]
 40%|████      | 12/30 [00:03<00:05,  3.34it/s]
 43%|████▎     | 13/30 [00:03<00:04,  3.42it/s]
 47%|████▋     | 14/30 [00:04<00:04,  3.59it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.57it/s]
 53%|█████▎    | 16/30 [00:04<00:03,  3.59it/s]
 57%|█████▋    | 17/30 [00:04<00:03,  3.45it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.45it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.53it/s]
 67%|██████▋   | 20/30 [00:05<00:02,  3.58it/s]
 70%|███████   | 21/30 [00:06<00:02,  3.66it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.73it/s]
 77%|███████▋  | 23/30 [00:06<00:01,  3.51it/s]
 80%|████████  | 24/30 [00:06<00:01,  3.60it/s]
 83%|████████▎ | 25/30 [00:07<00:01,  3.63it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.65it/s]
 90%|█████████ | 27/30 [00:07<00:00,  3.61it/s]
 93%|█████████▎| 28/30 [00:08<00:00,  3.62it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.64it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.71it/s]
 20%|██        | 2/10 [00:00<00:02,  3.47it/s]
 30%|███       | 3/10 [00:00<00:02,  3.44it/s]
 40%|████      | 4/10 [00:01<00:01,  3.47it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.51it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.48it/s]
 70%|███████   | 7/10 [00:01<00:00,  3.52it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.55it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.57it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.73it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.70it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.65it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.55it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.57it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.43it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.66it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.62it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.59it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.59it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.56it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.65it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:08,  3.78it/s]
  6%|▋         | 2/32 [00:00<00:08,  3.67it/s]
  9%|▉         | 3/32 [00:00<00:08,  3.58it/s]
 12%|█▎        | 4/32 [00:01<00:07,  3.57it/s]
 16%|█▌        | 5/32 [00:01<00:07,  3.49it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.45it/s]
 22%|██▏       | 7/32 [00:01<00:07,  3.50it/s]
 25%|██▌       | 8/32 [00:02<00:06,  3.48it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.53it/s]
 31%|███▏      | 10/32 [00:02<00:06,  3.62it/s]
 34%|███▍      | 11/32 [00:03<00:05,  3.66it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.67it/s]
 41%|████      | 13/32 [00:03<00:05,  3.66it/s]
 44%|████▍     | 14/32 [00:03<00:04,  3.65it/s]
 47%|████▋     | 15/32 [00:04<00:04,  3.69it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.70it/s]
 53%|█████▎    | 17/32 [00:04<00:03,  3.77it/s]
 56%|█████▋    | 18/32 [00:04<00:03,  3.78it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.74it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.64it/s]
 66%|██████▌   | 21/32 [00:05<00:03,  3.61it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.71it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.73it/s]
 75%|███████▌  | 24/32 [00:06<00:02,  3.71it/s]
 78%|███████▊  | 25/32 [00:06<00:01,  3.61it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.69it/s]
 84%|████████▍ | 27/32 [00:07<00:01,  3.68it/s]
 88%|████████▊ | 28/32 [00:07<00:01,  3.66it/s]
 91%|█████████ | 29/32 [00:07<00:00,  3.67it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.69it/s]
 97%|█████████▋| 31/32 [00:08<00:00,  3.64it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:05,  3.48it/s]
 11%|█         | 2/19 [00:00<00:04,  3.68it/s]
 16%|█▌        | 3/19 [00:00<00:04,  3.70it/s]
 21%|██        | 4/19 [00:01<00:04,  3.70it/s]
 26%|██▋       | 5/19 [00:01<00:03,  3.65it/s]
 32%|███▏      | 6/19 [00:01<00:03,  3.70it/s]
 37%|███▋      | 7/19 [00:01<00:03,  3.76it/s]
 42%|████▏     | 8/19 [00:02<00:02,  3.70it/s]
 47%|████▋     | 9/19 [00:02<00:02,  3.70it/s]
 53%|█████▎    | 10/19 [00:02<00:02,  3.65it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.61it/s]
 63%|██████▎   | 12/19 [00:03<00:01,  3.64it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.62it/s]
 74%|███████▎  | 14/19 [00:03<00:01,  3.59it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.65it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.67it/s]
 89%|████████▉ | 17/19 [00:04<00:00,  3.72it/s]
 95%|█████████▍| 18/19 [00:04<00:00,  3.65it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.47it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.65it/s]
 11%|█         | 3/27 [00:00<00:06,  3.68it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.63it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.65it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.66it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.63it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.60it/s]
 33%|███▎      | 9/27 [00:02<00:04,  3.64it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.62it/s]
 41%|████      | 11/27 [00:03<00:04,  3.59it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.55it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.60it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.59it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.66it/s]
 59%|█████▉    | 16/27 [00:04<00:02,  3.69it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.61it/s]
 67%|██████▋   | 18/27 [00:04<00:02,  3.58it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.56it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.65it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.65it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.69it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.66it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.68it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.62it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.65it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.66it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.59it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.59it/s]
 31%|███       | 4/13 [00:01<00:02,  3.61it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.61it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.62it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.61it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.48it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.45it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.43it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.44it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.48it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.49it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.52it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.16it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.26it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.39it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.41it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.48it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.52it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.54it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.45it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.54it/s]
 12%|█▎        | 3/24 [00:00<00:05,  3.65it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.56it/s]
 21%|██        | 5/24 [00:01<00:05,  3.55it/s]
 25%|██▌       | 6/24 [00:01<00:05,  3.58it/s]
 29%|██▉       | 7/24 [00:01<00:04,  3.61it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.64it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.65it/s]
 42%|████▏     | 10/24 [00:02<00:03,  3.63it/s]
 46%|████▌     | 11/24 [00:03<00:03,  3.64it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.64it/s]
 54%|█████▍    | 13/24 [00:03<00:02,  3.67it/s]
 58%|█████▊    | 14/24 [00:03<00:02,  3.61it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.62it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.62it/s]
 71%|███████   | 17/24 [00:04<00:01,  3.60it/s]
 75%|███████▌  | 18/24 [00:04<00:01,  3.60it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.57it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.62it/s]
 88%|████████▊ | 21/24 [00:05<00:00,  3.61it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.60it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.57it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.19it/s]
 22%|██▏       | 2/9 [00:00<00:02,  3.32it/s]
 33%|███▎      | 3/9 [00:00<00:02,  2.96it/s]
 44%|████▍     | 4/9 [00:01<00:01,  2.87it/s]
 56%|████��▌    | 5/9 [00:01<00:01,  3.07it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.19it/s]
 78%|███████▊  | 7/9 [00:02<00:00,  3.35it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.43it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.40it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.54it/s]
 11%|█         | 3/27 [00:00<00:06,  3.59it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.61it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.62it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.73it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.69it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.69it/s]
 33%|███▎      | 9/27 [00:02<00:04,  3.67it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.70it/s]
 41%|████      | 11/27 [00:03<00:04,  3.65it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.64it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.66it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.61it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.61it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.63it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.67it/s]
 67%|██████▋   | 18/27 [00:04<00:02,  3.66it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.61it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.62it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.60it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.57it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.59it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.61it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.59it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.58it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  3.61it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.54it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.52it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.51it/s]
 42%|████▏     | 5/12 [00:01<00:01,  3.63it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.66it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.69it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.67it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.69it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.66it/s]
 92%|█████████▏| 11/12 [00:03<00:00,  3.68it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.63it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.68it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.66it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.72it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.77it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.69it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.66it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.63it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.58it/s]
 31%|███       | 4/13 [00:01<00:02,  3.56it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.59it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.60it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.59it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.65it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.67it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.64it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.67it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.55it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.48it/s]
 10%|█         | 2/20 [00:00<00:05,  3.57it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.61it/s]
 20%|██        | 4/20 [00:01<00:04,  3.66it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.72it/s]
 30%|███       | 6/20 [00:01<00:03,  3.73it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.70it/s]
 40%|████      | 8/20 [00:02<00:03,  3.79it/s]
 45%|████▌     | 9/20 [00:02<00:02,  3.80it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.74it/s]
 55%|█████▌    | 11/20 [00:02<00:02,  3.77it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.74it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.79it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.80it/s]
 75%|███████▌  | 15/20 [00:04<00:01,  3.75it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.80it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.72it/s]
 90%|█████████ | 18/20 [00:04<00:00,  3.74it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.70it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:03,  3.58it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.77it/s]
 20%|██        | 3/15 [00:00<00:03,  3.73it/s]
 27%|██▋       | 4/15 [00:01<00:02,  3.76it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.74it/s]
 40%|████      | 6/15 [00:01<00:02,  3.81it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.72it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.63it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.58it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.58it/s]
 73%|███████▎  | 11/15 [00:03<00:01,  3.60it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.63it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.66it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.70it/s]
+
+[2025-07-11 17:15:54] Weighted Per image PCK0.10: 51.32%, image PCK0.05: 36.14%, image PCK0.01: 60.80%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.5_layer_5_no_dino.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:08:15] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.5_layer_5_no_dino', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.5_layer_5'], NO_DINO=True, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='7', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.5_layer_5']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:20,  1.44s/it]
 13%|█▎        | 2/15 [00:01<00:11,  1.10it/s]
 20%|██        | 3/15 [00:02<00:07,  1.57it/s]
 27%|██▋       | 4/15 [00:02<00:05,  2.01it/s]
 33%|███▎      | 5/15 [00:02<00:04,  2.38it/s]
 40%|████      | 6/15 [00:03<00:03,  2.64it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.85it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  3.04it/s]
 60%|██████    | 9/15 [00:04<00:01,  3.20it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.31it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.38it/s]
 80%|████████  | 12/15 [00:04<00:00,  3.43it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.49it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.49it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:07,  3.78it/s]
  7%|▋         | 2/30 [00:00<00:07,  3.82it/s]
 10%|█         | 3/30 [00:00<00:07,  3.64it/s]
 13%|█▎        | 4/30 [00:01<00:07,  3.31it/s]
 17%|█▋        | 5/30 [00:01<00:07,  3.52it/s]
 20%|██        | 6/30 [00:01<00:06,  3.59it/s]
 23%|██▎       | 7/30 [00:01<00:06,  3.64it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.46it/s]
 30%|███       | 9/30 [00:02<00:05,  3.57it/s]
 33%|███▎      | 10/30 [00:02<00:05,  3.65it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.63it/s]
 40%|████      | 12/30 [00:03<00:06,  2.87it/s]
 43%|████▎     | 13/30 [00:04<00:06,  2.70it/s]
 47%|████▋     | 14/30 [00:04<00:06,  2.59it/s]
 50%|█████     | 15/30 [00:04<00:05,  2.83it/s]
 53%|█████▎    | 16/30 [00:04<00:04,  3.09it/s]
 57%|█████▋    | 17/30 [00:05<00:04,  3.19it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.35it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.47it/s]
 67%|██████▋   | 20/30 [00:06<00:02,  3.55it/s]
 70%|███████   | 21/30 [00:06<00:02,  3.15it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  2.98it/s]
 77%|███████▋  | 23/30 [00:07<00:02,  2.84it/s]
 80%|████████  | 24/30 [00:07<00:02,  2.98it/s]
 83%|████████▎ | 25/30 [00:07<00:01,  3.18it/s]
 87%|████████▋ | 26/30 [00:08<00:01,  3.32it/s]
 90%|█████████ | 27/30 [00:08<00:00,  3.17it/s]
 93%|█████████▎| 28/30 [00:08<00:00,  3.32it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.46it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.57it/s]
 20%|██        | 2/10 [00:00<00:02,  3.58it/s]
 30%|███       | 3/10 [00:00<00:01,  3.69it/s]
 40%|████      | 4/10 [00:01<00:01,  3.67it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.50it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.65it/s]
 70%|███████   | 7/10 [00:01<00:00,  3.70it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.18it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.18it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:02,  2.42it/s]
 33%|███▎      | 2/6 [00:00<00:01,  2.99it/s]
 50%|█████     | 3/6 [00:01<00:01,  2.78it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.07it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.30it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:01,  3.75it/s]
 25%|██▌       | 2/8 [00:00<00:02,  2.92it/s]
 38%|███▊      | 3/8 [00:01<00:01,  2.85it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.07it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.34it/s]
 75%|███████▌  | 6/8 [00:02<00:00,  2.84it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.03it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:13,  2.37it/s]
  6%|▋         | 2/32 [00:00<00:12,  2.37it/s]
  9%|▉         | 3/32 [00:01<00:10,  2.73it/s]
 12%|█▎        | 4/32 [00:01<00:09,  3.08it/s]
 16%|█▌        | 5/32 [00:01<00:08,  3.29it/s]
 19%|█▉        | 6/32 [00:02<00:09,  2.65it/s]
 22%|██▏       | 7/32 [00:02<00:08,  2.87it/s]
 25%|██▌       | 8/32 [00:02<00:09,  2.48it/s]
 28%|██▊       | 9/32 [00:03<00:08,  2.75it/s]
 31%|███▏      | 10/32 [00:03<00:07,  2.96it/s]
 34%|███▍      | 11/32 [00:03<00:06,  3.14it/s]
 38%|███▊      | 12/32 [00:04<00:06,  3.26it/s]
 41%|████      | 13/32 [00:04<00:05,  3.39it/s]
 44%|████▍     | 14/32 [00:04<00:05,  3.53it/s]
 47%|████▋     | 15/32 [00:04<00:04,  3.56it/s]
 50%|█████     | 16/32 [00:05<00:04,  3.58it/s]
 53%|█████▎    | 17/32 [00:05<00:04,  3.59it/s]
 56%|█████▋    | 18/32 [00:05<00:03,  3.64it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.67it/s]
 62%|██████▎   | 20/32 [00:06<00:03,  3.70it/s]
 66%|██████▌   | 21/32 [00:06<00:02,  3.75it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.71it/s]
 72%|███████▏  | 23/32 [00:07<00:02,  3.65it/s]
 75%|███████▌  | 24/32 [00:07<00:02,  3.71it/s]
 78%|███████▊  | 25/32 [00:07<00:01,  3.80it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.78it/s]
 84%|████████▍ | 27/32 [00:08<00:01,  3.84it/s]
 88%|████████▊ | 28/32 [00:08<00:01,  3.71it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.69it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.71it/s]
 97%|█████████▋| 31/32 [00:09<00:00,  3.82it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:09,  1.80it/s]
 11%|█         | 2/19 [00:00<00:06,  2.53it/s]
 16%|█▌        | 3/19 [00:01<00:05,  3.00it/s]
 21%|██        | 4/19 [00:01<00:04,  3.30it/s]
 26%|██▋       | 5/19 [00:01<00:04,  3.34it/s]
 32%|███▏      | 6/19 [00:01<00:03,  3.41it/s]
 37%|███▋      | 7/19 [00:02<00:04,  2.80it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.03it/s]
 47%|████▋     | 9/19 [00:02<00:03,  3.19it/s]
 53%|█████▎    | 10/19 [00:03<00:02,  3.38it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.50it/s]
 63%|██████▎   | 12/19 [00:03<00:01,  3.56it/s]
 68%|██████▊   | 13/19 [00:04<00:01,  3.64it/s]
 74%|███████▎  | 14/19 [00:04<00:01,  3.71it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.70it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.73it/s]
 89%|████████▉ | 17/19 [00:05<00:00,  3.72it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.77it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:13,  1.94it/s]
  7%|▋         | 2/27 [00:00<00:09,  2.69it/s]
 11%|█         | 3/27 [00:01<00:07,  3.04it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.34it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.45it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.53it/s]
 26%|██▌       | 7/27 [00:02<00:05,  3.53it/s]
 30%|██▉       | 8/27 [00:02<00:06,  3.07it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.12it/s]
 37%|███▋      | 10/27 [00:03<00:06,  2.70it/s]
 41%|████      | 11/27 [00:03<00:05,  2.87it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.12it/s]
 48%|████▊     | 13/27 [00:04<00:04,  3.24it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.38it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.45it/s]
 59%|█████▉    | 16/27 [00:05<00:03,  3.49it/s]
 63%|██████▎   | 17/27 [00:05<00:02,  3.55it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.62it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.63it/s]
 74%|███████▍  | 20/27 [00:06<00:01,  3.59it/s]
 78%|███████▊  | 21/27 [00:06<00:01,  3.61it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.62it/s]
 85%|████████▌ | 23/27 [00:07<00:01,  3.00it/s]
 89%|████████▉ | 24/27 [00:07<00:00,  3.14it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  2.73it/s]
 96%|█████████▋| 26/27 [00:08<00:00,  2.88it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.72it/s]
 15%|█▌        | 2/13 [00:00<00:04,  2.59it/s]
 23%|██▎       | 3/13 [00:01<00:03,  2.89it/s]
 31%|███       | 4/13 [00:01<00:03,  2.68it/s]
 38%|███▊      | 5/13 [00:01<00:02,  2.75it/s]
 46%|████▌     | 6/13 [00:02<00:02,  2.96it/s]
 54%|█████▍    | 7/13 [00:02<00:01,  3.14it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.33it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.45it/s]
 77%|███████▋  | 10/13 [00:03<00:00,  3.53it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.62it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.48it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.49it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  2.57it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  2.75it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.18it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.29it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.52it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.59it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.63it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.68it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:09,  2.41it/s]
  8%|▊         | 2/24 [00:00<00:07,  3.08it/s]
 12%|█▎        | 3/24 [00:00<00:06,  3.09it/s]
 17%|█▋        | 4/24 [00:01<00:06,  3.30it/s]
 21%|██        | 5/24 [00:01<00:05,  3.31it/s]
 25%|██▌       | 6/24 [00:01<00:05,  3.43it/s]
 29%|██▉       | 7/24 [00:02<00:04,  3.45it/s]
 33%|███▎      | 8/24 [00:02<00:05,  3.10it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.12it/s]
 42%|████▏     | 10/24 [00:03<00:04,  3.07it/s]
 46%|████▌     | 11/24 [00:03<00:04,  3.21it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.37it/s]
 54%|█████▍    | 13/24 [00:04<00:03,  3.19it/s]
 58%|█████▊    | 14/24 [00:04<00:03,  3.08it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.12it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.28it/s]
 71%|███████   | 17/24 [00:05<00:02,  3.43it/s]
 75%|███████▌  | 18/24 [00:05<00:01,  3.41it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.52it/s]
 83%|████████▎ | 20/24 [00:06<00:01,  3.56it/s]
 88%|████████▊ | 21/24 [00:06<00:00,  3.62it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.33it/s]
 96%|█████████▌| 23/24 [00:07<00:00,  3.17it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.73it/s]
 22%|██▏       | 2/9 [00:00<00:01,  3.79it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.84it/s]
 44%|████▍     | 4/9 [00:01<00:01,  2.92it/s]
 56%|█████��    | 5/9 [00:01<00:01,  3.03it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.22it/s]
 78%|███████▊  | 7/9 [00:02<00:00,  3.35it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.26it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:09,  2.79it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.22it/s]
 11%|█         | 3/27 [00:00<00:06,  3.47it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.53it/s]
 19%|█▊        | 5/27 [00:01<00:07,  2.85it/s]
 22%|██▏       | 6/27 [00:01<00:06,  3.06it/s]
 26%|██▌       | 7/27 [00:02<00:06,  3.28it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.39it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.46it/s]
 37%|███▋      | 10/27 [00:03<00:04,  3.52it/s]
 41%|████      | 11/27 [00:03<00:04,  3.58it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.64it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.70it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.34it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.39it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.49it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.58it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.66it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.75it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.76it/s]
 78%|███████▊  | 21/27 [00:06<00:01,  3.44it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.32it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.20it/s]
 89%|████████▉ | 24/27 [00:07<00:00,  3.38it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.53it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.57it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:02,  3.72it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.88it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.86it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.90it/s]
 42%|████▏     | 5/12 [00:01<00:01,  3.80it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.82it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.75it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.76it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.67it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.67it/s]
 92%|█████████▏| 11/12 [00:02<00:00,  3.56it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:02,  2.76it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.18it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.39it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.45it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.40it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.35it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.66it/s]
 15%|█▌        | 2/13 [00:00<00:02,  3.68it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.72it/s]
 31%|███       | 4/13 [00:01<00:02,  3.42it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.50it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.55it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.47it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.52it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.57it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.64it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.57it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.65it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.61it/s]
 10%|█         | 2/20 [00:00<00:05,  3.45it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.56it/s]
 20%|██        | 4/20 [00:01<00:04,  3.48it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.51it/s]
 30%|███       | 6/20 [00:01<00:03,  3.60it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.69it/s]
 40%|████      | 8/20 [00:02<00:03,  3.63it/s]
 45%|████▌     | 9/20 [00:02<00:03,  3.64it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.68it/s]
 55%|█████▌    | 11/20 [00:03<00:02,  3.69it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.70it/s]
 65%|██████▌   | 13/20 [00:03<00:02,  3.43it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.52it/s]
 75%|███████▌  | 15/20 [00:04<00:01,  3.63it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.64it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.66it/s]
 90%|█████████ | 18/20 [00:05<00:00,  3.42it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.54it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:04,  2.92it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.25it/s]
 20%|██        | 3/15 [00:00<00:03,  3.44it/s]
 27%|██▋       | 4/15 [00:01<00:03,  3.51it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.58it/s]
 40%|████      | 6/15 [00:01<00:02,  3.53it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.62it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.65it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.53it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.56it/s]
 73%|███████▎  | 11/15 [00:03<00:01,  3.61it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.61it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.61it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.62it/s]
+
+[2025-07-11 17:09:47] Weighted Per image PCK0.10: 45.04%, image PCK0.05: 20.32%, image PCK0.01: 63.74%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.6_layer_10.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:17:27] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.6_layer_10', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.6_layer_10'], NO_DINO=False, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='4', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.6_layer_10']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:20,  1.50s/it]
 13%|█▎        | 2/15 [00:02<00:12,  1.04it/s]
 20%|██        | 3/15 [00:02<00:08,  1.46it/s]
 27%|██▋       | 4/15 [00:02<00:05,  1.84it/s]
 33%|███▎      | 5/15 [00:03<00:04,  2.18it/s]
 40%|████      | 6/15 [00:03<00:03,  2.51it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.73it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  2.91it/s]
 60%|██████    | 9/15 [00:04<00:01,  3.04it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.16it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.22it/s]
 80%|████████  | 12/15 [00:05<00:00,  3.24it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.31it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.34it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:08,  3.39it/s]
  7%|▋         | 2/30 [00:00<00:08,  3.39it/s]
 10%|█         | 3/30 [00:00<00:08,  3.27it/s]
 13%|█▎        | 4/30 [00:01<00:07,  3.32it/s]
 17%|█▋        | 5/30 [00:01<00:07,  3.29it/s]
 20%|██        | 6/30 [00:01<00:07,  3.29it/s]
 23%|██▎       | 7/30 [00:02<00:06,  3.34it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.44it/s]
 30%|███       | 9/30 [00:02<00:06,  3.44it/s]
 33%|███▎      | 10/30 [00:02<00:05,  3.49it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.46it/s]
 40%|████      | 12/30 [00:03<00:05,  3.46it/s]
 43%|████▎     | 13/30 [00:03<00:04,  3.48it/s]
 47%|████▋     | 14/30 [00:04<00:04,  3.54it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.54it/s]
 53%|█████▎    | 16/30 [00:04<00:03,  3.53it/s]
 57%|█████▋    | 17/30 [00:04<00:03,  3.50it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.52it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.55it/s]
 67%|██████▋   | 20/30 [00:05<00:02,  3.56it/s]
 70%|███████   | 21/30 [00:06<00:02,  3.60it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.65it/s]
 77%|███████▋  | 23/30 [00:06<00:01,  3.60it/s]
 80%|████████  | 24/30 [00:06<00:01,  3.64it/s]
 83%|████████▎ | 25/30 [00:07<00:01,  3.61it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.61it/s]
 90%|█████████ | 27/30 [00:07<00:00,  3.62it/s]
 93%|█████████▎| 28/30 [00:07<00:00,  3.59it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.59it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.01it/s]
 20%|██        | 2/10 [00:00<00:02,  3.22it/s]
 30%|███       | 3/10 [00:00<00:02,  3.34it/s]
 40%|████      | 4/10 [00:01<00:01,  3.42it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.49it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.47it/s]
 70%|███████   | 7/10 [00:02<00:00,  3.51it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.54it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.56it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.76it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.71it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.67it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.57it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.57it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:01,  3.55it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.68it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.60it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.57it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.58it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.56it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.65it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:08,  3.79it/s]
  6%|▋         | 2/32 [00:00<00:08,  3.66it/s]
  9%|▉         | 3/32 [00:00<00:08,  3.60it/s]
 12%|█▎        | 4/32 [00:01<00:07,  3.58it/s]
 16%|█▌        | 5/32 [00:01<00:07,  3.61it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.54it/s]
 22%|██▏       | 7/32 [00:01<00:07,  3.55it/s]
 25%|██▌       | 8/32 [00:02<00:06,  3.51it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.49it/s]
 31%|███▏      | 10/32 [00:02<00:06,  3.59it/s]
 34%|███▍      | 11/32 [00:03<00:05,  3.63it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.59it/s]
 41%|████      | 13/32 [00:03<00:05,  3.57it/s]
 44%|████▍     | 14/32 [00:03<00:05,  3.55it/s]
 47%|████▋     | 15/32 [00:04<00:04,  3.60it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.61it/s]
 53%|█████▎    | 17/32 [00:04<00:04,  3.66it/s]
 56%|█████▋    | 18/32 [00:05<00:03,  3.66it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.66it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.63it/s]
 66%|██████▌   | 21/32 [00:05<00:03,  3.61it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.68it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.68it/s]
 75%|███████▌  | 24/32 [00:06<00:02,  3.65it/s]
 78%|███████▊  | 25/32 [00:06<00:01,  3.59it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.64it/s]
 84%|████████▍ | 27/32 [00:07<00:01,  3.61it/s]
 88%|████████▊ | 28/32 [00:07<00:01,  3.64it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.67it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.65it/s]
 97%|█████████▋| 31/32 [00:08<00:00,  3.58it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:05,  3.40it/s]
 11%|█         | 2/19 [00:00<00:04,  3.47it/s]
 16%|█▌        | 3/19 [00:00<00:04,  3.53it/s]
 21%|██        | 4/19 [00:01<00:04,  3.17it/s]
 26%|██▋       | 5/19 [00:01<00:04,  3.24it/s]
 32%|███▏      | 6/19 [00:01<00:04,  3.19it/s]
 37%|███▋      | 7/19 [00:02<00:03,  3.35it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.40it/s]
 47%|████▋     | 9/19 [00:02<00:02,  3.45it/s]
 53%|█████▎    | 10/19 [00:02<00:02,  3.45it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.48it/s]
 63%|██████▎   | 12/19 [00:03<00:01,  3.52it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.48it/s]
 74%|███████▎  | 14/19 [00:04<00:01,  3.49it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.58it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.61it/s]
 89%|████████▉ | 17/19 [00:04<00:00,  3.65it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.58it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.55it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.62it/s]
 11%|█         | 3/27 [00:00<00:06,  3.69it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.61it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.62it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.63it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.62it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.57it/s]
 33%|███▎      | 9/27 [00:02<00:04,  3.62it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.61it/s]
 41%|████      | 11/27 [00:03<00:04,  3.60it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.55it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.62it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.59it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.62it/s]
 59%|█████▉    | 16/27 [00:04<00:02,  3.67it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.56it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.52it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.52it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.55it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.60it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.65it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.64it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.65it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.64it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.63it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.43it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.43it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.45it/s]
 31%|███       | 4/13 [00:01<00:02,  3.50it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.55it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.59it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.58it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.53it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.51it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.48it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.47it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.51it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.40it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.52it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.45it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.43it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.49it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.47it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.51it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.54it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.55it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.50it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.55it/s]
 12%|█▎        | 3/24 [00:00<00:05,  3.54it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.48it/s]
 21%|██        | 5/24 [00:01<00:05,  3.48it/s]
 25%|██▌       | 6/24 [00:01<00:05,  3.50it/s]
 29%|██▉       | 7/24 [00:01<00:04,  3.50it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.51it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.51it/s]
 42%|████▏     | 10/24 [00:02<00:04,  3.46it/s]
 46%|████▌     | 11/24 [00:03<00:03,  3.49it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.51it/s]
 54%|█████▍    | 13/24 [00:03<00:03,  3.53it/s]
 58%|█████▊    | 14/24 [00:03<00:02,  3.52it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.56it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.56it/s]
 71%|███████   | 17/24 [00:04<00:01,  3.56it/s]
 75%|███████▌  | 18/24 [00:05<00:01,  3.57it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.56it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.61it/s]
 88%|████████▊ | 21/24 [00:05<00:00,  3.58it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.56it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.54it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.46it/s]
 22%|██▏       | 2/9 [00:00<00:02,  3.47it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.42it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.39it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.44it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.45it/s]
 78%|███████▊  | 7/9 [00:02<00:00,  3.52it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.53it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.27it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.39it/s]
 11%|█         | 3/27 [00:00<00:06,  3.43it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.44it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.40it/s]
 22%|██▏       | 6/27 [00:01<00:06,  3.50it/s]
 26%|██▌       | 7/27 [00:02<00:05,  3.44it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.45it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.49it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.50it/s]
 41%|████      | 11/27 [00:03<00:04,  3.50it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.48it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.51it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.53it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.54it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.55it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.56it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.55it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.53it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.54it/s]
 78%|███████▊  | 21/27 [00:06<00:01,  3.54it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.51it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.54it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.54it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.51it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.53it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  3.58it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.50it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.48it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.43it/s]
 42%|████▏     | 5/12 [00:01<00:01,  3.54it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.56it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.61it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.59it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.61it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.59it/s]
 92%|█████████▏| 11/12 [00:03<00:00,  3.61it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.62it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.69it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.64it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.67it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.69it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.63it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.63it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.59it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.56it/s]
 31%|███       | 4/13 [00:01<00:02,  3.53it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.56it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.56it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.56it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.58it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.61it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.57it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.64it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.56it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.52it/s]
 10%|█         | 2/20 [00:00<00:05,  3.59it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.59it/s]
 20%|██        | 4/20 [00:01<00:04,  3.60it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.66it/s]
 30%|███       | 6/20 [00:01<00:03,  3.63it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.58it/s]
 40%|████      | 8/20 [00:02<00:03,  3.66it/s]
 45%|████▌     | 9/20 [00:02<00:02,  3.68it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.63it/s]
 55%|█████▌    | 11/20 [00:03<00:02,  3.68it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.65it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.69it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.69it/s]
 75%|███████▌  | 15/20 [00:04<00:01,  3.65it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.69it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.62it/s]
 90%|█████████ | 18/20 [00:04<00:00,  3.65it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.60it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:03,  3.61it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.68it/s]
 20%|██        | 3/15 [00:00<00:03,  3.61it/s]
 27%|██▋       | 4/15 [00:01<00:03,  3.60it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.57it/s]
 40%|████      | 6/15 [00:01<00:02,  3.64it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.57it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.55it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.54it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.53it/s]
 73%|███████▎  | 11/15 [00:03<00:01,  3.54it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.55it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.58it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.61it/s]
+
+[2025-07-11 17:18:55] Weighted Per image PCK0.10: 81.60%, image PCK0.05: 67.25%, image PCK0.01: 87.15%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.6_layer_10_no_dino.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:17:32] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.6_layer_10_no_dino', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.6_layer_10'], NO_DINO=True, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='5', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.6_layer_10']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:19,  1.43s/it]
 13%|█▎        | 2/15 [00:01<00:11,  1.09it/s]
 20%|██        | 3/15 [00:02<00:07,  1.60it/s]
 27%|██▋       | 4/15 [00:02<00:05,  2.04it/s]
 33%|███▎      | 5/15 [00:02<00:04,  2.42it/s]
 40%|████      | 6/15 [00:03<00:03,  2.69it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.99it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  3.20it/s]
 60%|██████    | 9/15 [00:03<00:01,  3.32it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.45it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.58it/s]
 80%|████████  | 12/15 [00:04<00:00,  3.63it/s]
 87%|████████▋ | 13/15 [00:04<00:00,  3.53it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.60it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:07,  3.89it/s]
  7%|▋         | 2/30 [00:00<00:08,  3.23it/s]
 10%|█         | 3/30 [00:00<00:08,  3.34it/s]
 13%|█▎        | 4/30 [00:01<00:07,  3.40it/s]
 17%|█▋        | 5/30 [00:01<00:07,  3.50it/s]
 20%|██        | 6/30 [00:01<00:06,  3.56it/s]
 23%|██▎       | 7/30 [00:02<00:06,  3.55it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.37it/s]
 30%|███       | 9/30 [00:02<00:06,  3.43it/s]
 33%|███▎      | 10/30 [00:02<00:05,  3.47it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.54it/s]
 40%|████      | 12/30 [00:03<00:05,  3.57it/s]
 43%|████▎     | 13/30 [00:03<00:04,  3.62it/s]
 47%|████▋     | 14/30 [00:03<00:04,  3.58it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.63it/s]
 53%|█████▎    | 16/30 [00:04<00:03,  3.69it/s]
 57%|█████▋    | 17/30 [00:04<00:03,  3.72it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.60it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.60it/s]
 67%|██████▋   | 20/30 [00:05<00:02,  3.61it/s]
 70%|███████   | 21/30 [00:05<00:02,  3.59it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.57it/s]
 77%|███████▋  | 23/30 [00:06<00:01,  3.69it/s]
 80%|████████  | 24/30 [00:06<00:01,  3.66it/s]
 83%|████████▎ | 25/30 [00:06<00:01,  3.68it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.73it/s]
 90%|█████████ | 27/30 [00:07<00:00,  3.74it/s]
 93%|█████████▎| 28/30 [00:07<00:00,  3.78it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.77it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.67it/s]
 20%|██        | 2/10 [00:00<00:02,  3.68it/s]
 30%|███       | 3/10 [00:00<00:01,  3.78it/s]
 40%|████      | 4/10 [00:01<00:01,  3.42it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.50it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.65it/s]
 70%|███████   | 7/10 [00:01<00:00,  3.69it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.70it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.67it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.18it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.39it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.15it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.37it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.54it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:01,  3.58it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.55it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.58it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.52it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.58it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.63it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.60it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:09,  3.12it/s]
  6%|▋         | 2/32 [00:00<00:09,  3.29it/s]
  9%|▉         | 3/32 [00:00<00:09,  3.15it/s]
 12%|█▎        | 4/32 [00:01<00:08,  3.31it/s]
 16%|█▌        | 5/32 [00:01<00:08,  3.32it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.31it/s]
 22%|██▏       | 7/32 [00:02<00:07,  3.39it/s]
 25%|██▌       | 8/32 [00:02<00:06,  3.49it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.52it/s]
 31%|███▏      | 10/32 [00:02<00:06,  3.48it/s]
 34%|███▍      | 11/32 [00:03<00:06,  3.48it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.46it/s]
 41%|████      | 13/32 [00:03<00:05,  3.50it/s]
 44%|████▍     | 14/32 [00:04<00:05,  3.49it/s]
 47%|████▋     | 15/32 [00:04<00:04,  3.52it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.52it/s]
 53%|█████▎    | 17/32 [00:04<00:04,  3.39it/s]
 56%|█████▋    | 18/32 [00:05<00:04,  3.43it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.41it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.45it/s]
 66%|██████▌   | 21/32 [00:06<00:03,  3.47it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.41it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.38it/s]
 75%|███████▌  | 24/32 [00:07<00:02,  3.44it/s]
 78%|███████▊  | 25/32 [00:07<00:02,  3.50it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.40it/s]
 84%|████████▍ | 27/32 [00:07<00:01,  3.45it/s]
 88%|████████▊ | 28/32 [00:08<00:01,  3.43it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.43it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.44it/s]
 97%|█████████▋| 31/32 [00:09<00:00,  3.52it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:05,  3.57it/s]
 11%|█         | 2/19 [00:00<00:04,  3.56it/s]
 16%|█▌        | 3/19 [00:00<00:04,  3.57it/s]
 21%|██        | 4/19 [00:01<00:04,  3.53it/s]
 26%|██▋       | 5/19 [00:01<00:04,  3.42it/s]
 32%|███▏      | 6/19 [00:01<00:03,  3.36it/s]
 37%|███▋      | 7/19 [00:02<00:03,  3.37it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.21it/s]
 47%|████▋     | 9/19 [00:02<00:03,  3.26it/s]
 53%|█████▎    | 10/19 [00:02<00:02,  3.37it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.45it/s]
 63%|██████▎   | 12/19 [00:03<00:02,  3.44it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.48it/s]
 74%|███████▎  | 14/19 [00:04<00:01,  3.53it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.44it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.43it/s]
 89%|████████▉ | 17/19 [00:04<00:00,  3.37it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.43it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:08,  2.93it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.23it/s]
 11%|█         | 3/27 [00:00<00:07,  3.34it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.49it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.51it/s]
 22%|██▏       | 6/27 [00:01<00:06,  3.48it/s]
 26%|██▌       | 7/27 [00:02<00:05,  3.51it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.52it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.52it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.55it/s]
 41%|████      | 11/27 [00:03<00:04,  3.58it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.61it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.57it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.60it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.57it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.56it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.59it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.64it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.64it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.58it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.58it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.55it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.57it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.57it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.60it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.57it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.47it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.57it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.48it/s]
 31%|███       | 4/13 [00:01<00:02,  3.32it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.40it/s]
 46%|████▌     | 6/13 [00:01<00:02,  3.35it/s]
 54%|█████▍    | 7/13 [00:02<00:01,  3.44it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.53it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.59it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.64it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.69it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.69it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.62it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.62it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  2.76it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.21it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.28it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.47it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.50it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.52it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.57it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.56it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.66it/s]
 12%|█▎        | 3/24 [00:00<00:05,  3.64it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.72it/s]
 21%|██        | 5/24 [00:01<00:05,  3.69it/s]
 25%|██▌       | 6/24 [00:01<00:04,  3.68it/s]
 29%|██▉       | 7/24 [00:01<00:04,  3.66it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.63it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.64it/s]
 42%|████▏     | 10/24 [00:02<00:04,  3.44it/s]
 46%|████▌     | 11/24 [00:03<00:03,  3.48it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.51it/s]
 54%|█████▍    | 13/24 [00:03<00:03,  3.54it/s]
 58%|█████▊    | 14/24 [00:03<00:02,  3.54it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.54it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.56it/s]
 71%|███████   | 17/24 [00:04<00:01,  3.61it/s]
 75%|███████▌  | 18/24 [00:05<00:01,  3.58it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.54it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.53it/s]
 88%|████████▊ | 21/24 [00:05<00:00,  3.53it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.56it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.58it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.55it/s]
 22%|██▏       | 2/9 [00:00<00:01,  3.62it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.69it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.78it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.75it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.66it/s]
 78%|███████▊  | 7/9 [00:01<00:00,  3.61it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.61it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.47it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.60it/s]
 11%|█         | 3/27 [00:00<00:06,  3.60it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.61it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.58it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.53it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.55it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.54it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.56it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.58it/s]
 41%|████      | 11/27 [00:03<00:04,  3.61it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.62it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.68it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.63it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.60it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.59it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.59it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.62it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.65it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.64it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.67it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.73it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.68it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.56it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.63it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.62it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  3.57it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.78it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.78it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.81it/s]
 42%|████▏     | 5/12 [00:01<00:01,  3.72it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.72it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.68it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.70it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.67it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.66it/s]
 92%|█████████▏| 11/12 [00:02<00:00,  3.62it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.45it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.49it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.58it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.54it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.53it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.58it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.62it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.67it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.67it/s]
 31%|███       | 4/13 [00:01<00:02,  3.71it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.68it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.68it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.69it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.67it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.66it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.72it/s]
 85%|████████▍ | 11/13 [00:02<00:00,  3.66it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.71it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.56it/s]
 10%|█         | 2/20 [00:00<00:04,  3.63it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.67it/s]
 20%|██        | 4/20 [00:01<00:04,  3.68it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.64it/s]
 30%|███       | 6/20 [00:01<00:03,  3.64it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.70it/s]
 40%|████      | 8/20 [00:02<00:03,  3.64it/s]
 45%|████▌     | 9/20 [00:02<00:03,  3.62it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.65it/s]
 55%|█████▌    | 11/20 [00:03<00:02,  3.64it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.66it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.61it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.61it/s]
 75%|███████▌  | 15/20 [00:04<00:01,  3.63it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.62it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.72it/s]
 90%|█████████ | 18/20 [00:04<00:00,  3.67it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.70it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:04,  3.44it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.48it/s]
 20%|██        | 3/15 [00:00<00:03,  3.57it/s]
 27%|██▋       | 4/15 [00:01<00:03,  3.58it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.63it/s]
 40%|████      | 6/15 [00:01<00:02,  3.62it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.71it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.73it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.73it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.74it/s]
 73%|███████▎  | 11/15 [00:02<00:01,  3.72it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.70it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.66it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.63it/s]
+
+[2025-07-11 17:18:58] Weighted Per image PCK0.10: 73.77%, image PCK0.05: 57.32%, image PCK0.01: 82.15%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.6_layer_11_no_dino.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:17:36] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.6_layer_11_no_dino', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.6_layer_11'], NO_DINO=True, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='7', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.6_layer_11']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:19,  1.39s/it]
 13%|█▎        | 2/15 [00:01<00:11,  1.10it/s]
 20%|██        | 3/15 [00:02<00:08,  1.49it/s]
 27%|██▋       | 4/15 [00:02<00:05,  1.92it/s]
 33%|███▎      | 5/15 [00:02<00:04,  2.31it/s]
 40%|████      | 6/15 [00:03<00:03,  2.59it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.87it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  3.03it/s]
 60%|██████    | 9/15 [00:04<00:01,  3.11it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.22it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.28it/s]
 80%|████████  | 12/15 [00:04<00:00,  3.41it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.47it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.47it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:08,  3.51it/s]
  7%|▋         | 2/30 [00:00<00:07,  3.57it/s]
 10%|█         | 3/30 [00:00<00:07,  3.50it/s]
 13%|█▎        | 4/30 [00:01<00:07,  3.51it/s]
 17%|█▋        | 5/30 [00:01<00:06,  3.65it/s]
 20%|██        | 6/30 [00:01<00:06,  3.72it/s]
 23%|██▎       | 7/30 [00:01<00:06,  3.68it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.63it/s]
 30%|███       | 9/30 [00:02<00:05,  3.65it/s]
 33%|███▎      | 10/30 [00:02<00:05,  3.65it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.70it/s]
 40%|████      | 12/30 [00:03<00:04,  3.71it/s]
 43%|████▎     | 13/30 [00:03<00:04,  3.72it/s]
 47%|████▋     | 14/30 [00:03<00:04,  3.68it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.69it/s]
 53%|█████▎    | 16/30 [00:04<00:03,  3.74it/s]
 57%|█████▋    | 17/30 [00:04<00:03,  3.79it/s]
 60%|██████    | 18/30 [00:04<00:03,  3.75it/s]
 63%|██████▎   | 19/30 [00:05<00:02,  3.72it/s]
 67%|██████▋   | 20/30 [00:05<00:02,  3.69it/s]
 70%|███████   | 21/30 [00:05<00:02,  3.66it/s]
 73%|███████▎  | 22/30 [00:05<00:02,  3.63it/s]
 77%|███████▋  | 23/30 [00:06<00:01,  3.71it/s]
 80%|████████  | 24/30 [00:06<00:01,  3.67it/s]
 83%|████████▎ | 25/30 [00:06<00:01,  3.69it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.69it/s]
 90%|█████████ | 27/30 [00:07<00:00,  3.67it/s]
 93%|█████████▎| 28/30 [00:07<00:00,  3.68it/s]
 97%|█████████▋| 29/30 [00:07<00:00,  3.67it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.58it/s]
 20%|██        | 2/10 [00:00<00:02,  3.59it/s]
 30%|███       | 3/10 [00:00<00:01,  3.68it/s]
 40%|████      | 4/10 [00:01<00:01,  3.66it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.64it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.72it/s]
 70%|███████   | 7/10 [00:01<00:00,  3.72it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.73it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.67it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.44it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.50it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.64it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.68it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.73it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:01,  3.72it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.71it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.74it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.74it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.78it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.78it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.73it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:08,  3.51it/s]
  6%|▋         | 2/32 [00:00<00:08,  3.63it/s]
  9%|▉         | 3/32 [00:00<00:07,  3.64it/s]
 12%|█▎        | 4/32 [00:01<00:07,  3.67it/s]
 16%|█▌        | 5/32 [00:01<00:07,  3.64it/s]
 19%|█▉        | 6/32 [00:01<00:06,  3.72it/s]
 22%|██▏       | 7/32 [00:01<00:06,  3.72it/s]
 25%|██▌       | 8/32 [00:02<00:06,  3.78it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.78it/s]
 31%|███▏      | 10/32 [00:02<00:05,  3.74it/s]
 34%|███▍      | 11/32 [00:02<00:05,  3.69it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.67it/s]
 41%|████      | 13/32 [00:03<00:05,  3.72it/s]
 44%|████▍     | 14/32 [00:03<00:04,  3.76it/s]
 47%|████▋     | 15/32 [00:04<00:04,  3.75it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.75it/s]
 53%|█████▎    | 17/32 [00:04<00:04,  3.69it/s]
 56%|█████▋    | 18/32 [00:04<00:03,  3.70it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.70it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.71it/s]
 66%|██████▌   | 21/32 [00:05<00:02,  3.73it/s]
 69%|██████▉   | 22/32 [00:05<00:02,  3.66it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.66it/s]
 75%|███████▌  | 24/32 [00:06<00:02,  3.71it/s]
 78%|███████▊  | 25/32 [00:06<00:01,  3.74it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.70it/s]
 84%|████████▍ | 27/32 [00:07<00:01,  3.71it/s]
 88%|████████▊ | 28/32 [00:07<00:01,  3.66it/s]
 91%|█████████ | 29/32 [00:07<00:00,  3.64it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.64it/s]
 97%|█████████▋| 31/32 [00:08<00:00,  3.66it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:04,  3.62it/s]
 11%|█         | 2/19 [00:00<00:04,  3.65it/s]
 16%|█▌        | 3/19 [00:00<00:04,  3.66it/s]
 21%|██        | 4/19 [00:01<00:04,  3.66it/s]
 26%|██▋       | 5/19 [00:01<00:03,  3.71it/s]
 32%|███▏      | 6/19 [00:01<00:03,  3.50it/s]
 37%|███▋      | 7/19 [00:02<00:03,  3.34it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.47it/s]
 47%|████▋     | 9/19 [00:02<00:02,  3.52it/s]
 53%|█████▎    | 10/19 [00:02<00:02,  3.60it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.64it/s]
 63%|██████▎   | 12/19 [00:03<00:01,  3.59it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.56it/s]
 74%|███████▎  | 14/19 [00:03<00:01,  3.59it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.59it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.61it/s]
 89%|████████▉ | 17/19 [00:04<00:00,  3.62it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.65it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:06,  3.72it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.76it/s]
 11%|█         | 3/27 [00:00<00:06,  3.66it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.69it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.66it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.63it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.66it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.66it/s]
 33%|███▎      | 9/27 [00:02<00:04,  3.65it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.67it/s]
 41%|████      | 11/27 [00:02<00:04,  3.68it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.72it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.67it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.69it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.68it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.66it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.68it/s]
 67%|██████▋   | 18/27 [00:04<00:02,  3.71it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.69it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.68it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.64it/s]
 81%|████████▏ | 22/27 [00:05<00:01,  3.63it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.64it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.63it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.66it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.67it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.60it/s]
 15%|█▌        | 2/13 [00:00<00:02,  3.78it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.86it/s]
 31%|███       | 4/13 [00:01<00:02,  3.84it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.85it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.73it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.75it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.83it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.89it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.94it/s]
 85%|████████▍ | 11/13 [00:02<00:00,  3.99it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.88it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.07it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.40it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.49it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.58it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.64it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.74it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.75it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.75it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.77it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.57it/s]
  8%|▊         | 2/24 [00:00<00:05,  3.75it/s]
 12%|█▎        | 3/24 [00:00<00:05,  3.74it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.85it/s]
 21%|██        | 5/24 [00:01<00:04,  3.81it/s]
 25%|██▌       | 6/24 [00:01<00:04,  3.79it/s]
 29%|██▉       | 7/24 [00:01<00:04,  3.79it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.78it/s]
 38%|███▊      | 9/24 [00:02<00:03,  3.78it/s]
 42%|████▏     | 10/24 [00:02<00:03,  3.82it/s]
 46%|████▌     | 11/24 [00:02<00:03,  3.82it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.82it/s]
 54%|█████▍    | 13/24 [00:03<00:02,  3.81it/s]
 58%|█████▊    | 14/24 [00:03<00:02,  3.80it/s]
 62%|██████▎   | 15/24 [00:03<00:02,  3.80it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.81it/s]
 71%|███████   | 17/24 [00:04<00:01,  3.85it/s]
 75%|███████▌  | 18/24 [00:04<00:01,  3.83it/s]
 79%|███████▉  | 19/24 [00:04<00:01,  3.83it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.79it/s]
 88%|████████▊ | 21/24 [00:05<00:00,  3.77it/s]
 92%|█████████▏| 22/24 [00:05<00:00,  3.79it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.82it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.67it/s]
 22%|██▏       | 2/9 [00:00<00:01,  3.69it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.81it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.86it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.79it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.76it/s]
 78%|███████▊  | 7/9 [00:01<00:00,  3.73it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.70it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.47it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.62it/s]
 11%|█         | 3/27 [00:00<00:06,  3.69it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.67it/s]
 19%|█▊        | 5/27 [00:01<00:05,  3.70it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.66it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.70it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.67it/s]
 33%|███▎      | 9/27 [00:02<00:04,  3.67it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.69it/s]
 41%|████      | 11/27 [00:02<00:04,  3.69it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.72it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.77it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.76it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.79it/s]
 59%|█████▉    | 16/27 [00:04<00:02,  3.76it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.74it/s]
 67%|██████▋   | 18/27 [00:04<00:02,  3.76it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.78it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.75it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.75it/s]
 81%|████████▏ | 22/27 [00:05<00:01,  3.81it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.79it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.82it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.88it/s]
 96%|█████████▋| 26/27 [00:06<00:00,  3.86it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  3.61it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.80it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.86it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.93it/s]
 42%|████▏     | 5/12 [00:01<00:01,  3.82it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.84it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.81it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.82it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.80it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.79it/s]
 92%|█████████▏| 11/12 [00:02<00:00,  3.76it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.53it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.59it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.67it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.67it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.66it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.71it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.74it/s]
 15%|█▌        | 2/13 [00:00<00:02,  3.78it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.79it/s]
 31%|███       | 4/13 [00:01<00:02,  3.81it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.76it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.77it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.78it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.75it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.73it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.79it/s]
 85%|████████▍ | 11/13 [00:02<00:00,  3.70it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.78it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.66it/s]
 10%|█         | 2/20 [00:00<00:04,  3.69it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.74it/s]
 20%|██        | 4/20 [00:01<00:04,  3.74it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.74it/s]
 30%|███       | 6/20 [00:01<00:03,  3.79it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.85it/s]
 40%|████      | 8/20 [00:02<00:03,  3.78it/s]
 45%|████▌     | 9/20 [00:02<00:02,  3.75it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.80it/s]
 55%|█████▌    | 11/20 [00:02<00:02,  3.79it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.82it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.79it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.79it/s]
 75%|███████▌  | 15/20 [00:03<00:01,  3.80it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.77it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.86it/s]
 90%|█████████ | 18/20 [00:04<00:00,  3.82it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.86it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:03,  3.50it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.60it/s]
 20%|██        | 3/15 [00:00<00:03,  3.67it/s]
 27%|██▋       | 4/15 [00:01<00:02,  3.68it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.74it/s]
 40%|████      | 6/15 [00:01<00:02,  3.71it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.83it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.86it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.84it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.85it/s]
 73%|███████▎  | 11/15 [00:02<00:01,  3.86it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.83it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.75it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.74it/s]
+
+[2025-07-11 17:18:59] Weighted Per image PCK0.10: 74.95%, image PCK0.05: 57.26%, image PCK0.01: 83.24%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.6_layer_12.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:18:59] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.6_layer_12', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.6_layer_12'], NO_DINO=False, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='4', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.6_layer_12']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:20,  1.47s/it]
 13%|█▎        | 2/15 [00:01<00:11,  1.11it/s]
 20%|██        | 3/15 [00:02<00:08,  1.47it/s]
 27%|██▋       | 4/15 [00:02<00:06,  1.78it/s]
 33%|███▎      | 5/15 [00:03<00:04,  2.08it/s]
 40%|████      | 6/15 [00:03<00:03,  2.41it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.70it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  2.91it/s]
 60%|██████    | 9/15 [00:04<00:01,  3.07it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.18it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.25it/s]
 80%|████████  | 12/15 [00:05<00:00,  3.30it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.36it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.42it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:08,  3.42it/s]
  7%|▋         | 2/30 [00:00<00:09,  3.09it/s]
 10%|█         | 3/30 [00:00<00:08,  3.15it/s]
 13%|█▎        | 4/30 [00:01<00:08,  3.21it/s]
 17%|█▋        | 5/30 [00:01<00:07,  3.13it/s]
 20%|██        | 6/30 [00:01<00:07,  3.21it/s]
 23%|██▎       | 7/30 [00:02<00:06,  3.30it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.41it/s]
 30%|███       | 9/30 [00:02<00:06,  3.43it/s]
 33%|███▎      | 10/30 [00:03<00:05,  3.46it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.44it/s]
 40%|████      | 12/30 [00:03<00:05,  3.43it/s]
 43%|████▎     | 13/30 [00:03<00:05,  3.29it/s]
 47%|████▋     | 14/30 [00:04<00:04,  3.41it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.44it/s]
 53%|█████▎    | 16/30 [00:04<00:04,  3.50it/s]
 57%|█████▋    | 17/30 [00:05<00:03,  3.46it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.35it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.39it/s]
 67%|██████▋   | 20/30 [00:05<00:02,  3.40it/s]
 70%|███████   | 21/30 [00:06<00:02,  3.46it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.52it/s]
 77%|███████▋  | 23/30 [00:06<00:02,  3.49it/s]
 80%|████████  | 24/30 [00:07<00:01,  3.52it/s]
 83%|████████▎ | 25/30 [00:07<00:01,  3.50it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.50it/s]
 90%|█████████ | 27/30 [00:07<00:00,  3.48it/s]
 93%|█████████▎| 28/30 [00:08<00:00,  3.47it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.46it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.29it/s]
 20%|██        | 2/10 [00:00<00:02,  3.41it/s]
 30%|███       | 3/10 [00:00<00:02,  3.39it/s]
 40%|████      | 4/10 [00:01<00:01,  3.43it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.45it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.41it/s]
 70%|███████   | 7/10 [00:02<00:00,  3.43it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.44it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.47it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.66it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.61it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.58it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.49it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.50it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.50it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.66it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.59it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.56it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.55it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.52it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.62it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:08,  3.54it/s]
  6%|▋         | 2/32 [00:00<00:08,  3.53it/s]
  9%|▉         | 3/32 [00:00<00:08,  3.50it/s]
 12%|█▎        | 4/32 [00:01<00:08,  3.47it/s]
 16%|█▌        | 5/32 [00:01<00:07,  3.50it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.45it/s]
 22%|██▏       | 7/32 [00:02<00:07,  3.47it/s]
 25%|██▌       | 8/32 [00:02<00:06,  3.45it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.45it/s]
 31%|███▏      | 10/32 [00:02<00:06,  3.52it/s]
 34%|███▍      | 11/32 [00:03<00:05,  3.55it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.55it/s]
 41%|████      | 13/32 [00:03<00:05,  3.50it/s]
 44%|████▍     | 14/32 [00:04<00:05,  3.48it/s]
 47%|████▋     | 15/32 [00:04<00:04,  3.51it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.51it/s]
 53%|█████▎    | 17/32 [00:04<00:04,  3.58it/s]
 56%|█████▋    | 18/32 [00:05<00:03,  3.58it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.57it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.54it/s]
 66%|██████▌   | 21/32 [00:05<00:03,  3.52it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.59it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.58it/s]
 75%|███████▌  | 24/32 [00:06<00:02,  3.54it/s]
 78%|███████▊  | 25/32 [00:07<00:02,  3.48it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.52it/s]
 84%|████████▍ | 27/32 [00:07<00:01,  3.50it/s]
 88%|████████▊ | 28/32 [00:07<00:01,  3.53it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.56it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.57it/s]
 97%|█████████▋| 31/32 [00:08<00:00,  3.51it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:05,  3.38it/s]
 11%|█         | 2/19 [00:00<00:04,  3.50it/s]
 16%|█▌        | 3/19 [00:00<00:04,  3.53it/s]
 21%|██        | 4/19 [00:01<00:04,  3.56it/s]
 26%|██▋       | 5/19 [00:01<00:03,  3.59it/s]
 32%|███▏      | 6/19 [00:01<00:03,  3.68it/s]
 37%|███▋      | 7/19 [00:01<00:03,  3.76it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.52it/s]
 47%|████▋     | 9/19 [00:02<00:02,  3.56it/s]
 53%|█████▎    | 10/19 [00:02<00:02,  3.50it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.52it/s]
 63%|██████▎   | 12/19 [00:03<00:01,  3.57it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.55it/s]
 74%|███████▎  | 14/19 [00:03<00:01,  3.56it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.65it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.65it/s]
 89%|████████▉ | 17/19 [00:04<00:00,  3.69it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.64it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.58it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.67it/s]
 11%|█         | 3/27 [00:00<00:06,  3.79it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.64it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.61it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.63it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.64it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.59it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.60it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.59it/s]
 41%|████      | 11/27 [00:03<00:04,  3.57it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.56it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.60it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.57it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.62it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.63it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.59it/s]
 67%|██████▋   | 18/27 [00:04<00:02,  3.57it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.57it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.66it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.70it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.75it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.73it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.74it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.74it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.71it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.66it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.52it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.53it/s]
 31%|███       | 4/13 [00:01<00:02,  3.54it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.55it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.61it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.61it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.57it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.54it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.53it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.55it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.62it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.59it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.64it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:01,  3.60it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.51it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.57it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.51it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.54it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.61it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.61it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.50it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.58it/s]
 12%|█▎        | 3/24 [00:00<00:05,  3.66it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.61it/s]
 21%|██        | 5/24 [00:01<00:05,  3.62it/s]
 25%|██▌       | 6/24 [00:01<00:05,  3.60it/s]
 29%|██▉       | 7/24 [00:01<00:04,  3.64it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.61it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.61it/s]
 42%|████▏     | 10/24 [00:02<00:04,  3.50it/s]
 46%|████▌     | 11/24 [00:03<00:03,  3.50it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.51it/s]
 54%|█████▍    | 13/24 [00:03<00:03,  3.53it/s]
 58%|█████▊    | 14/24 [00:03<00:02,  3.54it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.59it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.60it/s]
 71%|███████   | 17/24 [00:04<00:01,  3.57it/s]
 75%|███████▌  | 18/24 [00:05<00:01,  3.59it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.61it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.67it/s]
 88%|████████▊ | 21/24 [00:05<00:00,  3.67it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.67it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.65it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.54it/s]
 22%|██▏       | 2/9 [00:00<00:01,  3.57it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.58it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.56it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.62it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.68it/s]
 78%|███████▊  | 7/9 [00:01<00:00,  3.76it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.76it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.53it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.61it/s]
 11%|█         | 3/27 [00:00<00:06,  3.66it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.71it/s]
 19%|█▊        | 5/27 [00:01<00:05,  3.67it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.70it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.62it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.63it/s]
 33%|███▎      | 9/27 [00:02<00:04,  3.65it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.63it/s]
 41%|████      | 11/27 [00:03<00:04,  3.64it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.61it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.58it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.61it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.60it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.64it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.65it/s]
 67%|██████▋   | 18/27 [00:04<00:02,  3.63it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.63it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.66it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.65it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.62it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.60it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.60it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.57it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.55it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  3.65it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.56it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.55it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.51it/s]
 42%|████▏     | 5/12 [00:01<00:01,  3.62it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.64it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.70it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.69it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.71it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.67it/s]
 92%|█████████▏| 11/12 [00:03<00:00,  3.69it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.81it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.80it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.76it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.79it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.82it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.75it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.71it/s]
 15%|█▌        | 2/13 [00:00<00:02,  3.68it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.68it/s]
 31%|███       | 4/13 [00:01<00:02,  3.66it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.67it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.63it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.60it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.59it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.61it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.57it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.61it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.54it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.54it/s]
 10%|█         | 2/20 [00:00<00:04,  3.61it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.60it/s]
 20%|██        | 4/20 [00:01<00:04,  3.66it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.70it/s]
 30%|███       | 6/20 [00:01<00:03,  3.67it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.61it/s]
 40%|████      | 8/20 [00:02<00:03,  3.68it/s]
 45%|████▌     | 9/20 [00:02<00:02,  3.71it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.68it/s]
 55%|█████▌    | 11/20 [00:02<00:02,  3.75it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.71it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.77it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.78it/s]
 75%|███████▌  | 15/20 [00:04<00:01,  3.72it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.75it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.66it/s]
 90%|█████████ | 18/20 [00:04<00:00,  3.65it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.61it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:03,  3.72it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.84it/s]
 20%|██        | 3/15 [00:00<00:03,  3.78it/s]
 27%|██▋       | 4/15 [00:01<00:02,  3.79it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.74it/s]
 40%|████      | 6/15 [00:01<00:02,  3.78it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.71it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.63it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.63it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.61it/s]
 73%|███████▎  | 11/15 [00:02<00:01,  3.61it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.62it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.65it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.69it/s]
+
+[2025-07-11 17:20:26] Weighted Per image PCK0.10: 81.16%, image PCK0.05: 64.89%, image PCK0.01: 87.74%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.6_layer_13_no_dino.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:19:07] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.6_layer_13_no_dino', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.6_layer_13'], NO_DINO=True, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='7', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.6_layer_13']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:19,  1.39s/it]
 13%|█▎        | 2/15 [00:01<00:11,  1.13it/s]
 20%|██        | 3/15 [00:02<00:07,  1.63it/s]
 27%|██▋       | 4/15 [00:02<00:05,  2.08it/s]
 33%|███▎      | 5/15 [00:02<00:04,  2.40it/s]
 40%|████      | 6/15 [00:03<00:03,  2.69it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.87it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  3.00it/s]
 60%|██████    | 9/15 [00:03<00:01,  3.15it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.16it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.24it/s]
 80%|████████  | 12/15 [00:04<00:00,  3.30it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.37it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.42it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:10,  2.79it/s]
  7%|▋         | 2/30 [00:00<00:08,  3.14it/s]
 10%|█         | 3/30 [00:00<00:08,  3.36it/s]
 13%|█▎        | 4/30 [00:01<00:07,  3.34it/s]
 17%|█▋        | 5/30 [00:01<00:07,  3.35it/s]
 20%|██        | 6/30 [00:01<00:07,  3.36it/s]
 23%|██▎       | 7/30 [00:02<00:06,  3.31it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.43it/s]
 30%|███       | 9/30 [00:02<00:06,  3.40it/s]
 33%|███▎      | 10/30 [00:02<00:05,  3.47it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.41it/s]
 40%|████      | 12/30 [00:03<00:05,  3.45it/s]
 43%|████▎     | 13/30 [00:03<00:04,  3.45it/s]
 47%|████▋     | 14/30 [00:04<00:04,  3.44it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.45it/s]
 53%|█████▎    | 16/30 [00:04<00:04,  3.46it/s]
 57%|█████▋    | 17/30 [00:05<00:03,  3.45it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.41it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.46it/s]
 67%|██████▋   | 20/30 [00:05<00:02,  3.48it/s]
 70%|███████   | 21/30 [00:06<00:02,  3.45it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.47it/s]
 77%|███████▋  | 23/30 [00:06<00:02,  3.45it/s]
 80%|████████  | 24/30 [00:07<00:01,  3.46it/s]
 83%|████████▎ | 25/30 [00:07<00:01,  3.45it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.46it/s]
 90%|█████████ | 27/30 [00:07<00:00,  3.48it/s]
 93%|█████████▎| 28/30 [00:08<00:00,  3.40it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.42it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.29it/s]
 20%|██        | 2/10 [00:00<00:02,  3.44it/s]
 30%|███       | 3/10 [00:00<00:02,  3.41it/s]
 40%|████      | 4/10 [00:01<00:01,  3.45it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.44it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.41it/s]
 70%|███████   | 7/10 [00:02<00:00,  3.36it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.41it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.44it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.14it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.45it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.28it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.34it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.40it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  2.87it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.33it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.43it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.43it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.48it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.49it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.60it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:09,  3.38it/s]
  6%|▋         | 2/32 [00:00<00:08,  3.49it/s]
  9%|▉         | 3/32 [00:00<00:08,  3.50it/s]
 12%|█▎        | 4/32 [00:01<00:08,  3.46it/s]
 16%|█▌        | 5/32 [00:01<00:07,  3.52it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.46it/s]
 22%|██▏       | 7/32 [00:02<00:07,  3.51it/s]
 25%|██▌       | 8/32 [00:02<00:06,  3.50it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.51it/s]
 31%|███▏      | 10/32 [00:02<00:06,  3.59it/s]
 34%|███▍      | 11/32 [00:03<00:05,  3.62it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.63it/s]
 41%|████      | 13/32 [00:03<00:05,  3.59it/s]
 44%|████▍     | 14/32 [00:03<00:05,  3.59it/s]
 47%|████▋     | 15/32 [00:04<00:04,  3.48it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.50it/s]
 53%|█████▎    | 17/32 [00:04<00:04,  3.54it/s]
 56%|█████▋    | 18/32 [00:05<00:03,  3.62it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.54it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.52it/s]
 66%|██████▌   | 21/32 [00:05<00:03,  3.57it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.50it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.54it/s]
 75%|███████▌  | 24/32 [00:06<00:02,  3.56it/s]
 78%|███████▊  | 25/32 [00:07<00:01,  3.57it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.67it/s]
 84%|████████▍ | 27/32 [00:07<00:01,  3.60it/s]
 88%|████████▊ | 28/32 [00:07<00:01,  3.67it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.70it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.69it/s]
 97%|█████████▋| 31/32 [00:08<00:00,  3.62it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:07,  2.53it/s]
 11%|█         | 2/19 [00:00<00:05,  2.99it/s]
 16%|█▌        | 3/19 [00:01<00:05,  2.88it/s]
 21%|██        | 4/19 [00:01<00:04,  3.03it/s]
 26%|██▋       | 5/19 [00:01<00:04,  2.97it/s]
 32%|███▏      | 6/19 [00:02<00:04,  2.95it/s]
 37%|███▋      | 7/19 [00:02<00:03,  3.20it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.12it/s]
 47%|████▋     | 9/19 [00:02<00:03,  3.25it/s]
 53%|█████▎    | 10/19 [00:03<00:02,  3.22it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.36it/s]
 63%|██████▎   | 12/19 [00:03<00:02,  3.07it/s]
 68%|██████▊   | 13/19 [00:04<00:01,  3.18it/s]
 74%|███████▎  | 14/19 [00:04<00:01,  3.27it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.38it/s]
 84%|████████▍ | 16/19 [00:05<00:00,  3.22it/s]
 89%|████████▉ | 17/19 [00:05<00:00,  3.30it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.37it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.67it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.69it/s]
 11%|█         | 3/27 [00:00<00:07,  3.26it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.33it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.38it/s]
 22%|██▏       | 6/27 [00:01<00:06,  3.45it/s]
 26%|██▌       | 7/27 [00:02<00:05,  3.45it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.25it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.38it/s]
 37%|███▋      | 10/27 [00:02<00:05,  3.39it/s]
 41%|████      | 11/27 [00:03<00:04,  3.43it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.44it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.55it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.51it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.62it/s]
 59%|█████▉    | 16/27 [00:04<00:02,  3.70it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.67it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.61it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.64it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.67it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.65it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.68it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.66it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.65it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.62it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.56it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:04,  2.84it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.17it/s]
 23%|██▎       | 3/13 [00:00<00:03,  3.21it/s]
 31%|███       | 4/13 [00:01<00:02,  3.36it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.36it/s]
 46%|████▌     | 6/13 [00:01<00:02,  3.43it/s]
 54%|█████▍    | 7/13 [00:02<00:01,  3.46it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.49it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.50it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.51it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.52it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.56it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.45it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.58it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:01,  3.57it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.53it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.56it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.57it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.52it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.55it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.60it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.57it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.55it/s]
 12%|█▎        | 3/24 [00:00<00:06,  3.46it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.43it/s]
 21%|██        | 5/24 [00:01<00:05,  3.48it/s]
 25%|██▌       | 6/24 [00:01<00:05,  3.57it/s]
 29%|██▉       | 7/24 [00:01<00:04,  3.55it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.55it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.54it/s]
 42%|████▏     | 10/24 [00:02<00:03,  3.51it/s]
 46%|████▌     | 11/24 [00:03<00:03,  3.53it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.53it/s]
 54%|█████▍    | 13/24 [00:03<00:03,  3.57it/s]
 58%|█████▊    | 14/24 [00:03<00:02,  3.59it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.62it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.64it/s]
 71%|███████   | 17/24 [00:04<00:01,  3.56it/s]
 75%|███████▌  | 18/24 [00:05<00:01,  3.61it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.60it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.56it/s]
 88%|████████▊ | 21/24 [00:05<00:00,  3.51it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.50it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.52it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.57it/s]
 22%|██▏       | 2/9 [00:00<00:01,  3.68it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.51it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.53it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.56it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.60it/s]
 78%|███████▊  | 7/9 [00:01<00:00,  3.69it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.67it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.65it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.64it/s]
 11%|█         | 3/27 [00:00<00:06,  3.64it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.62it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.61it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.69it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.64it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.66it/s]
 33%|███▎      | 9/27 [00:02<00:04,  3.69it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.71it/s]
 41%|████      | 11/27 [00:03<00:04,  3.68it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.65it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.65it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.64it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.65it/s]
 59%|█████▉    | 16/27 [00:04<00:02,  3.67it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.66it/s]
 67%|██████▋   | 18/27 [00:04<00:02,  3.64it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.64it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.65it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.64it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.59it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.60it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.58it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.50it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.51it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  3.44it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.42it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.42it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.42it/s]
 42%|████▏     | 5/12 [00:01<00:02,  3.48it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.53it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.61it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.57it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.59it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.56it/s]
 92%|█████████▏| 11/12 [00:03<00:00,  3.58it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.74it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.72it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.68it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.70it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.71it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.64it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.48it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.52it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.52it/s]
 31%|███       | 4/13 [00:01<00:02,  3.51it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.56it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.56it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.56it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.60it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.63it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.58it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.64it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.59it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.63it/s]
 10%|█         | 2/20 [00:00<00:04,  3.69it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.68it/s]
 20%|██        | 4/20 [00:01<00:04,  3.68it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.75it/s]
 30%|███       | 6/20 [00:01<00:03,  3.73it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.70it/s]
 40%|████      | 8/20 [00:02<00:03,  3.78it/s]
 45%|████▌     | 9/20 [00:02<00:02,  3.82it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.77it/s]
 55%|█████▌    | 11/20 [00:02<00:02,  3.81it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.76it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.81it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.83it/s]
 75%|███████▌  | 15/20 [00:03<00:01,  3.78it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.83it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.76it/s]
 90%|█████████ | 18/20 [00:04<00:00,  3.76it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.73it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:04,  3.32it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.59it/s]
 20%|██        | 3/15 [00:00<00:03,  3.62it/s]
 27%|██▋       | 4/15 [00:01<00:03,  3.65it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.67it/s]
 40%|████      | 6/15 [00:01<00:02,  3.77it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.70it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.66it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.64it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.64it/s]
 73%|███████▎  | 11/15 [00:03<00:01,  3.64it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.66it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.68it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.71it/s]
+
+[2025-07-11 17:20:34] Weighted Per image PCK0.10: 74.04%, image PCK0.05: 56.82%, image PCK0.01: 82.38%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.6_layer_15_no_dino.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:20:36] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.6_layer_15_no_dino', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.6_layer_15'], NO_DINO=True, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='5', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.6_layer_15']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:20,  1.44s/it]
 13%|█▎        | 2/15 [00:01<00:10,  1.24it/s]
 20%|██        | 3/15 [00:02<00:07,  1.57it/s]
 27%|██▋       | 4/15 [00:02<00:05,  1.90it/s]
 33%|███▎      | 5/15 [00:02<00:04,  2.29it/s]
 40%|████      | 6/15 [00:03<00:03,  2.59it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.74it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  2.93it/s]
 60%|██████    | 9/15 [00:04<00:01,  3.07it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.21it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.27it/s]
 80%|████████  | 12/15 [00:04<00:00,  3.24it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.34it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.38it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:08,  3.29it/s]
  7%|▋         | 2/30 [00:00<00:08,  3.47it/s]
 10%|█         | 3/30 [00:00<00:07,  3.45it/s]
 13%|█▎        | 4/30 [00:01<00:07,  3.47it/s]
 17%|█▋        | 5/30 [00:01<00:07,  3.49it/s]
 20%|██        | 6/30 [00:01<00:06,  3.52it/s]
 23%|██▎       | 7/30 [00:02<00:06,  3.52it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.38it/s]
 30%|███       | 9/30 [00:02<00:06,  3.43it/s]
 33%|███▎      | 10/30 [00:02<00:05,  3.45it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.48it/s]
 40%|████      | 12/30 [00:03<00:05,  3.45it/s]
 43%|████▎     | 13/30 [00:03<00:04,  3.42it/s]
 47%|████▋     | 14/30 [00:04<00:04,  3.40it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.43it/s]
 53%|█████▎    | 16/30 [00:04<00:04,  3.47it/s]
 57%|█████▋    | 17/30 [00:04<00:03,  3.53it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.52it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.52it/s]
 67%|██████▋   | 20/30 [00:05<00:02,  3.51it/s]
 70%|███████   | 21/30 [00:06<00:02,  3.49it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.44it/s]
 77%|███████▋  | 23/30 [00:06<00:01,  3.53it/s]
 80%|████████  | 24/30 [00:06<00:01,  3.41it/s]
 83%|████████▎ | 25/30 [00:07<00:01,  3.42it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.29it/s]
 90%|█████████ | 27/30 [00:07<00:00,  3.32it/s]
 93%|█████████▎| 28/30 [00:08<00:00,  3.40it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.44it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.01it/s]
 20%|██        | 2/10 [00:00<00:02,  3.16it/s]
 30%|███       | 3/10 [00:00<00:02,  3.42it/s]
 40%|████      | 4/10 [00:01<00:01,  3.48it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.50it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.59it/s]
 70%|███████   | 7/10 [00:02<00:00,  3.61it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.62it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.57it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.22it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.36it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.28it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.44it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.40it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:01,  3.66it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.62it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.64it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.65it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.66it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.70it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.66it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:09,  3.37it/s]
  6%|▋         | 2/32 [00:00<00:08,  3.51it/s]
  9%|▉         | 3/32 [00:00<00:08,  3.48it/s]
 12%|█▎        | 4/32 [00:01<00:07,  3.50it/s]
 16%|█▌        | 5/32 [00:01<00:07,  3.52it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.49it/s]
 22%|██▏       | 7/32 [00:01<00:07,  3.51it/s]
 25%|██▌       | 8/32 [00:02<00:06,  3.58it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.55it/s]
 31%|███▏      | 10/32 [00:02<00:06,  3.41it/s]
 34%|███▍      | 11/32 [00:03<00:06,  3.35it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.37it/s]
 41%|████      | 13/32 [00:03<00:05,  3.45it/s]
 44%|████▍     | 14/32 [00:04<00:05,  3.51it/s]
 47%|████▋     | 15/32 [00:04<00:05,  3.34it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.42it/s]
 53%|█████▎    | 17/32 [00:04<00:04,  3.42it/s]
 56%|█████▋    | 18/32 [00:05<00:04,  3.48it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.49it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.53it/s]
 66%|██████▌   | 21/32 [00:06<00:03,  3.51it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.44it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.41it/s]
 75%|███████▌  | 24/32 [00:06<00:02,  3.48it/s]
 78%|███████▊  | 25/32 [00:07<00:01,  3.54it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.55it/s]
 84%|████████▍ | 27/32 [00:07<00:01,  3.58it/s]
 88%|████████▊ | 28/32 [00:08<00:01,  3.53it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.49it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.41it/s]
 97%|█████████▋| 31/32 [00:08<00:00,  3.46it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:06,  2.90it/s]
 11%|█         | 2/19 [00:00<00:05,  3.24it/s]
 16%|█▌        | 3/19 [00:00<00:04,  3.35it/s]
 21%|██        | 4/19 [00:01<00:04,  3.27it/s]
 26%|██▋       | 5/19 [00:01<00:04,  3.37it/s]
 32%|███▏      | 6/19 [00:01<00:03,  3.40it/s]
 37%|███▋      | 7/19 [00:02<00:03,  3.42it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.49it/s]
 47%|████▋     | 9/19 [00:02<00:02,  3.48it/s]
 53%|█████▎    | 10/19 [00:02<00:02,  3.52it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.57it/s]
 63%|██████▎   | 12/19 [00:03<00:01,  3.57it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.61it/s]
 74%|███████▎  | 14/19 [00:04<00:01,  3.62it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.54it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.55it/s]
 89%|████████▉ | 17/19 [00:04<00:00,  3.49it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.55it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:08,  2.94it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.13it/s]
 11%|█         | 3/27 [00:00<00:07,  3.32it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.43it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.43it/s]
 22%|██▏       | 6/27 [00:01<00:06,  3.48it/s]
 26%|██▌       | 7/27 [00:02<00:05,  3.54it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.49it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.35it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.44it/s]
 41%|████      | 11/27 [00:03<00:04,  3.51it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.58it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.51it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.56it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.52it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.54it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.52it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.45it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.36it/s]
 74%|███████▍  | 20/27 [00:05<00:02,  3.38it/s]
 78%|███████▊  | 21/27 [00:06<00:01,  3.35it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.41it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.34it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.41it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.47it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.50it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:04,  2.84it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.32it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.43it/s]
 31%|███       | 4/13 [00:01<00:02,  3.24it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.40it/s]
 46%|████▌     | 6/13 [00:01<00:02,  3.40it/s]
 54%|█████▍    | 7/13 [00:02<00:01,  3.48it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.59it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.64it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.59it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.63it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.62it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.28it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.54it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:01,  3.66it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.75it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.76it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.73it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.66it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.64it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.65it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.58it/s]
  8%|▊         | 2/24 [00:00<00:05,  3.70it/s]
 12%|█▎        | 3/24 [00:00<00:05,  3.66it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.73it/s]
 21%|██        | 5/24 [00:01<00:05,  3.70it/s]
 25%|██▌       | 6/24 [00:01<00:04,  3.68it/s]
 29%|██▉       | 7/24 [00:01<00:04,  3.68it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.66it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.65it/s]
 42%|████▏     | 10/24 [00:02<00:03,  3.71it/s]
 46%|████▌     | 11/24 [00:02<00:03,  3.70it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.69it/s]
 54%|█████▍    | 13/24 [00:03<00:03,  3.66it/s]
 58%|█████▊    | 14/24 [00:03<00:02,  3.61it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.61it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.61it/s]
 71%|███████   | 17/24 [00:04<00:01,  3.61it/s]
 75%|███████▌  | 18/24 [00:04<00:01,  3.62it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.65it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.61it/s]
 88%|████████▊ | 21/24 [00:05<00:00,  3.64it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.65it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.67it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.56it/s]
 22%|██▏       | 2/9 [00:00<00:01,  3.65it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.70it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.76it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.73it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.70it/s]
 78%|███████▊  | 7/9 [00:01<00:00,  3.66it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.67it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:06,  3.80it/s]
  7%|▋         | 2/27 [00:00<00:06,  3.75it/s]
 11%|█         | 3/27 [00:00<00:06,  3.70it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.66it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.66it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.60it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.65it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.63it/s]
 33%|███▎      | 9/27 [00:02<00:04,  3.62it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.65it/s]
 41%|████      | 11/27 [00:03<00:04,  3.65it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.66it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.70it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.69it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.69it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.67it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.66it/s]
 67%|██████▋   | 18/27 [00:04<00:02,  3.67it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.67it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.66it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.67it/s]
 81%|████████▏ | 22/27 [00:05<00:01,  3.74it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.73it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.76it/s]
 93%|█████████▎| 25/27 [00:06<00:00,  3.78it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.75it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  3.37it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.63it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.74it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.85it/s]
 42%|████▏     | 5/12 [00:01<00:01,  3.75it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.78it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.73it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.76it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.74it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.73it/s]
 92%|█████████▏| 11/12 [00:02<00:00,  3.70it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:02,  2.90it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.17it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.28it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.34it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.37it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.48it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.07it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.34it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.49it/s]
 31%|███       | 4/13 [00:01<00:02,  3.57it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.59it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.63it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.65it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.53it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.54it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.60it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.58it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.67it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.56it/s]
 10%|█         | 2/20 [00:00<00:04,  3.61it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.67it/s]
 20%|██        | 4/20 [00:01<00:04,  3.68it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.63it/s]
 30%|███       | 6/20 [00:01<00:03,  3.66it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.70it/s]
 40%|████      | 8/20 [00:02<00:03,  3.64it/s]
 45%|████▌     | 9/20 [00:02<00:03,  3.64it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.68it/s]
 55%|█████▌    | 11/20 [00:03<00:02,  3.67it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.69it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.67it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.66it/s]
 75%|███████▌  | 15/20 [00:04<00:01,  3.71it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.70it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.78it/s]
 90%|█████████ | 18/20 [00:04<00:00,  3.73it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.76it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:05,  2.67it/s]
 13%|█▎        | 2/15 [00:00<00:04,  3.05it/s]
 20%|██        | 3/15 [00:00<00:03,  3.34it/s]
 27%|██▋       | 4/15 [00:01<00:03,  3.45it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.57it/s]
 40%|████      | 6/15 [00:01<00:02,  3.59it/s]
 47%|████▋     | 7/15 [00:02<00:02,  3.70it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.76it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.78it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.80it/s]
 73%|███████▎  | 11/15 [00:03<00:01,  3.80it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.79it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.75it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.71it/s]
+
+[2025-07-11 17:22:03] Weighted Per image PCK0.10: 67.90%, image PCK0.05: 52.07%, image PCK0.01: 74.99%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.6_layer_18.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:20:36] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.6_layer_18', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.6_layer_18'], NO_DINO=False, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='6', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.6_layer_18']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:20,  1.43s/it]
 13%|█▎        | 2/15 [00:01<00:09,  1.33it/s]
 20%|██        | 3/15 [00:02<00:07,  1.55it/s]
 27%|██▋       | 4/15 [00:02<00:05,  1.99it/s]
 33%|███▎      | 5/15 [00:02<00:04,  2.32it/s]
 40%|████      | 6/15 [00:03<00:03,  2.63it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.82it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  3.01it/s]
 60%|██████    | 9/15 [00:03<00:01,  3.18it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.28it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.28it/s]
 80%|████████  | 12/15 [00:04<00:00,  3.31it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.35it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.43it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:08,  3.43it/s]
  7%|▋         | 2/30 [00:00<00:09,  2.89it/s]
 10%|█         | 3/30 [00:00<00:08,  3.20it/s]
 13%|█▎        | 4/30 [00:01<00:08,  3.24it/s]
 17%|█▋        | 5/30 [00:01<00:07,  3.29it/s]
 20%|██        | 6/30 [00:01<00:07,  3.31it/s]
 23%|██▎       | 7/30 [00:02<00:07,  3.21it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.37it/s]
 30%|███       | 9/30 [00:02<00:06,  3.37it/s]
 33%|███▎      | 10/30 [00:03<00:05,  3.46it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.45it/s]
 40%|████      | 12/30 [00:03<00:05,  3.48it/s]
 43%|████▎     | 13/30 [00:03<00:04,  3.44it/s]
 47%|████▋     | 14/30 [00:04<00:04,  3.55it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.53it/s]
 53%|█████▎    | 16/30 [00:04<00:03,  3.52it/s]
 57%|█████▋    | 17/30 [00:05<00:03,  3.49it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.51it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.54it/s]
 67%|██████▋   | 20/30 [00:05<00:02,  3.55it/s]
 70%|███████   | 21/30 [00:06<00:02,  3.58it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.60it/s]
 77%|███████▋  | 23/30 [00:06<00:01,  3.55it/s]
 80%|████████  | 24/30 [00:06<00:01,  3.60it/s]
 83%|████████▎ | 25/30 [00:07<00:01,  3.56it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.55it/s]
 90%|█████████ | 27/30 [00:07<00:00,  3.56it/s]
 93%|█████████▎| 28/30 [00:08<00:00,  3.52it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.53it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.52it/s]
 20%|██        | 2/10 [00:00<00:02,  3.41it/s]
 30%|███       | 3/10 [00:00<00:02,  3.40it/s]
 40%|████      | 4/10 [00:01<00:01,  3.46it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.46it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.44it/s]
 70%|███████   | 7/10 [00:02<00:00,  3.47it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.48it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.50it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.60it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.62it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.56it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.50it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.38it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.43it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.58it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.54it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.53it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.55it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.53it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.63it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:08,  3.78it/s]
  6%|▋         | 2/32 [00:00<00:08,  3.59it/s]
  9%|▉         | 3/32 [00:00<00:08,  3.38it/s]
 12%|█▎        | 4/32 [00:01<00:08,  3.39it/s]
 16%|█▌        | 5/32 [00:01<00:07,  3.45it/s]
 19%|█▉        | 6/32 [00:01<00:07,  3.38it/s]
 22%|██▏       | 7/32 [00:02<00:07,  3.43it/s]
 25%|██▌       | 8/32 [00:02<00:07,  3.41it/s]
 28%|██▊       | 9/32 [00:02<00:06,  3.43it/s]
 31%|███▏      | 10/32 [00:02<00:06,  3.48it/s]
 34%|███▍      | 11/32 [00:03<00:05,  3.53it/s]
 38%|███▊      | 12/32 [00:03<00:05,  3.54it/s]
 41%|████      | 13/32 [00:03<00:05,  3.52it/s]
 44%|████▍     | 14/32 [00:04<00:05,  3.49it/s]
 47%|████▋     | 15/32 [00:04<00:04,  3.51it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.52it/s]
 53%|█████▎    | 17/32 [00:04<00:04,  3.59it/s]
 56%|█████▋    | 18/32 [00:05<00:03,  3.58it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.58it/s]
 62%|██████▎   | 20/32 [00:05<00:03,  3.56it/s]
 66%|██████▌   | 21/32 [00:05<00:03,  3.55it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.64it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.65it/s]
 75%|███████▌  | 24/32 [00:06<00:02,  3.62it/s]
 78%|███████▊  | 25/32 [00:07<00:01,  3.56it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.63it/s]
 84%|████████▍ | 27/32 [00:07<00:01,  3.60it/s]
 88%|████████▊ | 28/32 [00:07<00:01,  3.61it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.62it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.64it/s]
 97%|█████████▋| 31/32 [00:08<00:00,  3.57it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:05,  3.40it/s]
 11%|█         | 2/19 [00:00<00:05,  3.30it/s]
 16%|█▌        | 3/19 [00:00<00:04,  3.34it/s]
 21%|██        | 4/19 [00:01<00:04,  3.10it/s]
 26%|██▋       | 5/19 [00:01<00:04,  3.19it/s]
 32%|███▏      | 6/19 [00:01<00:03,  3.35it/s]
 37%|███▋      | 7/19 [00:02<00:03,  3.45it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.37it/s]
 47%|████▋     | 9/19 [00:02<00:02,  3.40it/s]
 53%|█████▎    | 10/19 [00:02<00:02,  3.37it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.42it/s]
 63%|██████▎   | 12/19 [00:03<00:02,  3.47it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.48it/s]
 74%|███████▎  | 14/19 [00:04<00:01,  3.49it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.58it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.61it/s]
 89%|████████▉ | 17/19 [00:04<00:00,  3.65it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.60it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.41it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.50it/s]
 11%|█         | 3/27 [00:00<00:06,  3.58it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.32it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.35it/s]
 22%|██▏       | 6/27 [00:01<00:06,  3.40it/s]
 26%|██▌       | 7/27 [00:02<00:05,  3.42it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.45it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.51it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.54it/s]
 41%|████      | 11/27 [00:03<00:04,  3.51it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.49it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.57it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.57it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.62it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.65it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.59it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.54it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.54it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.58it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.59it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.61it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.60it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.59it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.56it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.54it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.25it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.33it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.37it/s]
 31%|███       | 4/13 [00:01<00:02,  3.42it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.46it/s]
 46%|████▌     | 6/13 [00:01<00:02,  3.49it/s]
 54%|█████▍    | 7/13 [00:02<00:01,  3.49it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.46it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.46it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.43it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.43it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.48it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.47it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.52it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.49it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.47it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.51it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.47it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.51it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.52it/s]
 88%|████████▊ | 7/8 [00:01<00:00,  3.53it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.46it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.52it/s]
 12%|█▎        | 3/24 [00:00<00:05,  3.57it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.51it/s]
 21%|██        | 5/24 [00:01<00:05,  3.53it/s]
 25%|██▌       | 6/24 [00:01<00:05,  3.55it/s]
 29%|██▉       | 7/24 [00:01<00:04,  3.57it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.58it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.57it/s]
 42%|████▏     | 10/24 [00:02<00:03,  3.53it/s]
 46%|████▌     | 11/24 [00:03<00:03,  3.51it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.50it/s]
 54%|█████▍    | 13/24 [00:03<00:03,  3.52it/s]
 58%|█████▊    | 14/24 [00:03<00:02,  3.53it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.55it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.55it/s]
 71%|███████   | 17/24 [00:04<00:01,  3.54it/s]
 75%|███████▌  | 18/24 [00:05<00:01,  3.53it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.52it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.59it/s]
 88%|████████▊ | 21/24 [00:05<00:00,  3.57it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.54it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.53it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.49it/s]
 22%|██▏       | 2/9 [00:00<00:01,  3.53it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.50it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.48it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.52it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.54it/s]
 78%|███████▊  | 7/9 [00:01<00:00,  3.62it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.61it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.38it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.47it/s]
 11%|█         | 3/27 [00:00<00:06,  3.49it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.51it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.50it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.58it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.51it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.52it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.55it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.55it/s]
 41%|████      | 11/27 [00:03<00:04,  3.53it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.52it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.52it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.53it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.54it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.58it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.59it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.57it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.54it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.55it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.53it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.48it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.49it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.48it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.44it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.45it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  3.55it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.46it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.43it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.40it/s]
 42%|████▏     | 5/12 [00:01<00:02,  3.49it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.52it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.57it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.56it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.57it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.55it/s]
 92%|█████████▏| 11/12 [00:03<00:00,  3.57it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.58it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.63it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.62it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.68it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.71it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.63it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.56it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.57it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.58it/s]
 31%|███       | 4/13 [00:01<00:02,  3.53it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.54it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.55it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.54it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.56it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.59it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.55it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.60it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.54it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.62it/s]
 10%|█         | 2/20 [00:00<00:04,  3.64it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.61it/s]
 20%|██        | 4/20 [00:01<00:04,  3.61it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.66it/s]
 30%|███       | 6/20 [00:01<00:03,  3.62it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.58it/s]
 40%|████      | 8/20 [00:02<00:03,  3.65it/s]
 45%|████▌     | 9/20 [00:02<00:02,  3.67it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.62it/s]
 55%|█████▌    | 11/20 [00:03<00:02,  3.67it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.63it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.66it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.67it/s]
 75%|███████▌  | 15/20 [00:04<00:01,  3.66it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.70it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.63it/s]
 90%|█████████ | 18/20 [00:04<00:00,  3.64it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.61it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:03,  3.54it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.68it/s]
 20%|██        | 3/15 [00:00<00:03,  3.63it/s]
 27%|██▋       | 4/15 [00:01<00:02,  3.67it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.67it/s]
 40%|████      | 6/15 [00:01<00:02,  3.71it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.61it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.56it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.54it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.52it/s]
 73%|███████▎  | 11/15 [00:03<00:01,  3.51it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.53it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.56it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.60it/s]
+
+[2025-07-11 17:22:04] Weighted Per image PCK0.10: 66.51%, image PCK0.05: 50.80%, image PCK0.01: 72.97%

Code/Baselines/logs_pascal_zs_dit_2025-07-11_17-01-52/pascal_zs_dit_t_0.6_layer_20.log

@@ -0,0 +1,44 @@
+[2025-07-11 17:22:02] Namespace(config='./GeoAware-SC/configs/eval_zero_shot_pascal.yaml', SEED=42, NOTE='pascal_zs_dit_t_0.6_layer_20', SAMPLE=None, TEST_SAMPLE=0, TOTAL_SAVE_RESULT=0, IMG_THRESHOLD=False, ANNO_SIZE=840, LR=0.00125, WD=0.001, BZ=1, SCHEDULER=None, SCHEDULER_P1=0.3, EPOCH=1, EVAL_EPOCH=5000, NOT_WANDB=True, TRAIN_DATASET='spair', SD_FEATURE_KEYS=['t_0.6_layer_20'], NO_DINO=False, LOAD=None, DENSE_OBJ=1, GAUSSIAN_AUGMENT=0.1, FEAT_MAP_DROPOUT=0.2, ENSEMBLE=1, PROJ_DIM=768, PAIR_AUGMENT=False, SELF_CONTRAST_WEIGHT=0, SOFT_TRAIN_WINDOW=0, DO_EVAL=True, DUMMY_NET=True, EVAL_DATASET='pascal', AP10K_EVAL_SUBSET='intra-species', COMPUTE_GEOAWARE_METRICS=False, KPT_RESULT=False, ADAPT_FLIP=False, MUTUAL_NN=False, SOFT_EVAL=True, SOFT_EVAL_WINDOW=5, GPU_ID='4', NUM_PATCHES=60, BBOX_THRE=False, AUGMENT_FLIP=0, AUGMENT_DOUBLE_FLIP=0, AUGMENT_SELF_FLIP=0)
+preparing sd features of sd_feature_keys: ['t_0.6_layer_20']
+Number of parameters corr net: 120
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:01<00:22,  1.60s/it]
 13%|█▎        | 2/15 [00:02<00:12,  1.04it/s]
 20%|██        | 3/15 [00:02<00:08,  1.47it/s]
 27%|██▋       | 4/15 [00:02<00:05,  1.90it/s]
 33%|███▎      | 5/15 [00:03<00:04,  2.26it/s]
 40%|████      | 6/15 [00:03<00:03,  2.60it/s]
 47%|████▋     | 7/15 [00:03<00:02,  2.80it/s]
 53%|█████▎    | 8/15 [00:03<00:02,  2.98it/s]
 60%|██████    | 9/15 [00:04<00:01,  3.11it/s]
 67%|██████▋   | 10/15 [00:04<00:01,  3.20it/s]
 73%|███████▎  | 11/15 [00:04<00:01,  3.24it/s]
 80%|████████  | 12/15 [00:05<00:00,  3.28it/s]
 87%|████████▋ | 13/15 [00:05<00:00,  3.32it/s]
 93%|█████████▎| 14/15 [00:05<00:00,  3.37it/s]
+
+
  0%|          | 0/30 [00:00<?, ?it/s]
  3%|▎         | 1/30 [00:00<00:08,  3.39it/s]
  7%|▋         | 2/30 [00:00<00:08,  3.40it/s]
 10%|█         | 3/30 [00:00<00:07,  3.52it/s]
 13%|█▎        | 4/30 [00:01<00:07,  3.50it/s]
 17%|█▋        | 5/30 [00:01<00:07,  3.46it/s]
 20%|██        | 6/30 [00:01<00:06,  3.44it/s]
 23%|██▎       | 7/30 [00:02<00:06,  3.45it/s]
 27%|██▋       | 8/30 [00:02<00:06,  3.50it/s]
 30%|███       | 9/30 [00:02<00:06,  3.49it/s]
 33%|███▎      | 10/30 [00:02<00:05,  3.53it/s]
 37%|███▋      | 11/30 [00:03<00:05,  3.47it/s]
 40%|████      | 12/30 [00:03<00:05,  3.44it/s]
 43%|████▎     | 13/30 [00:03<00:04,  3.42it/s]
 47%|████▋     | 14/30 [00:04<00:04,  3.50it/s]
 50%|█████     | 15/30 [00:04<00:04,  3.52it/s]
 53%|█████▎    | 16/30 [00:04<00:03,  3.52it/s]
 57%|█████▋    | 17/30 [00:04<00:03,  3.42it/s]
 60%|██████    | 18/30 [00:05<00:03,  3.41it/s]
 63%|██████▎   | 19/30 [00:05<00:03,  3.42it/s]
 67%|██████▋   | 20/30 [00:05<00:02,  3.45it/s]
 70%|███████   | 21/30 [00:06<00:02,  3.50it/s]
 73%|███████▎  | 22/30 [00:06<00:02,  3.53it/s]
 77%|███████▋  | 23/30 [00:06<00:02,  3.47it/s]
 80%|████████  | 24/30 [00:06<00:01,  3.52it/s]
 83%|████████▎ | 25/30 [00:07<00:01,  3.49it/s]
 87%|████████▋ | 26/30 [00:07<00:01,  3.48it/s]
 90%|█████████ | 27/30 [00:07<00:00,  3.43it/s]
 93%|█████████▎| 28/30 [00:08<00:00,  3.43it/s]
 97%|█████████▋| 29/30 [00:08<00:00,  3.45it/s]
+
+
  0%|          | 0/10 [00:00<?, ?it/s]
 10%|█         | 1/10 [00:00<00:02,  3.36it/s]
 20%|██        | 2/10 [00:00<00:02,  3.42it/s]
 30%|███       | 3/10 [00:00<00:02,  3.41it/s]
 40%|████      | 4/10 [00:01<00:01,  3.41it/s]
 50%|█████     | 5/10 [00:01<00:01,  3.41it/s]
 60%|██████    | 6/10 [00:01<00:01,  3.36it/s]
 70%|███████   | 7/10 [00:02<00:00,  3.39it/s]
 80%|████████  | 8/10 [00:02<00:00,  3.41it/s]
 90%|█████████ | 9/10 [00:02<00:00,  3.42it/s]
+
+
  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:01,  3.58it/s]
 33%|███▎      | 2/6 [00:00<00:01,  3.56it/s]
 50%|█████     | 3/6 [00:00<00:00,  3.51it/s]
 67%|██████▋   | 4/6 [00:01<00:00,  3.43it/s]
 83%|████████▎ | 5/6 [00:01<00:00,  3.40it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.42it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.58it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.46it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.32it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.36it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.30it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.38it/s]
+
+
  0%|          | 0/32 [00:00<?, ?it/s]
  3%|▎         | 1/32 [00:00<00:08,  3.45it/s]
  6%|▋         | 2/32 [00:00<00:08,  3.36it/s]
  9%|▉         | 3/32 [00:00<00:08,  3.32it/s]
 12%|█▎        | 4/32 [00:01<00:08,  3.22it/s]
 16%|█▌        | 5/32 [00:01<00:08,  3.18it/s]
 19%|█▉        | 6/32 [00:01<00:08,  3.16it/s]
 22%|██▏       | 7/32 [00:02<00:07,  3.20it/s]
 25%|██▌       | 8/32 [00:02<00:07,  3.11it/s]
 28%|██▊       | 9/32 [00:02<00:07,  3.03it/s]
 31%|███▏      | 10/32 [00:03<00:06,  3.17it/s]
 34%|███▍      | 11/32 [00:03<00:06,  3.24it/s]
 38%|███▊      | 12/32 [00:03<00:06,  3.29it/s]
 41%|████      | 13/32 [00:04<00:05,  3.26it/s]
 44%|████▍     | 14/32 [00:04<00:05,  3.24it/s]
 47%|████▋     | 15/32 [00:04<00:05,  3.31it/s]
 50%|█████     | 16/32 [00:04<00:04,  3.35it/s]
 53%|█████▎    | 17/32 [00:05<00:04,  3.43it/s]
 56%|█████▋    | 18/32 [00:05<00:04,  3.47it/s]
 59%|█████▉    | 19/32 [00:05<00:03,  3.48it/s]
 62%|██████▎   | 20/32 [00:06<00:03,  3.47it/s]
 66%|██████▌   | 21/32 [00:06<00:03,  3.49it/s]
 69%|██████▉   | 22/32 [00:06<00:02,  3.59it/s]
 72%|███████▏  | 23/32 [00:06<00:02,  3.60it/s]
 75%|███████▌  | 24/32 [00:07<00:02,  3.56it/s]
 78%|███████▊  | 25/32 [00:07<00:01,  3.52it/s]
 81%|████████▏ | 26/32 [00:07<00:01,  3.56it/s]
 84%|████████▍ | 27/32 [00:08<00:01,  3.51it/s]
 88%|████████▊ | 28/32 [00:08<00:01,  3.52it/s]
 91%|█████████ | 29/32 [00:08<00:00,  3.52it/s]
 94%|█████████▍| 30/32 [00:08<00:00,  3.51it/s]
 97%|█████████▋| 31/32 [00:09<00:00,  3.45it/s]
+
+
  0%|          | 0/19 [00:00<?, ?it/s]
  5%|▌         | 1/19 [00:00<00:05,  3.26it/s]
 11%|█         | 2/19 [00:00<00:04,  3.40it/s]
 16%|█▌        | 3/19 [00:00<00:04,  3.42it/s]
 21%|██        | 4/19 [00:01<00:04,  3.42it/s]
 26%|██▋       | 5/19 [00:01<00:04,  3.38it/s]
 32%|███▏      | 6/19 [00:01<00:03,  3.46it/s]
 37%|███▋      | 7/19 [00:02<00:03,  3.54it/s]
 42%|████▏     | 8/19 [00:02<00:03,  3.47it/s]
 47%|████▋     | 9/19 [00:02<00:02,  3.48it/s]
 53%|█████▎    | 10/19 [00:02<00:02,  3.42it/s]
 58%|█████▊    | 11/19 [00:03<00:02,  3.43it/s]
 63%|██████▎   | 12/19 [00:03<00:02,  3.47it/s]
 68%|██████▊   | 13/19 [00:03<00:01,  3.45it/s]
 74%|███████▎  | 14/19 [00:04<00:01,  3.44it/s]
 79%|███████▉  | 15/19 [00:04<00:01,  3.49it/s]
 84%|████████▍ | 16/19 [00:04<00:00,  3.50it/s]
 89%|████████▉ | 17/19 [00:04<00:00,  3.53it/s]
 95%|█████████▍| 18/19 [00:05<00:00,  3.48it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.41it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.50it/s]
 11%|█         | 3/27 [00:00<00:06,  3.55it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.47it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.47it/s]
 22%|██▏       | 6/27 [00:01<00:06,  3.49it/s]
 26%|██▌       | 7/27 [00:02<00:05,  3.48it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.47it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.49it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.47it/s]
 41%|████      | 11/27 [00:03<00:04,  3.46it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.42it/s]
 48%|████▊     | 13/27 [00:03<00:04,  3.48it/s]
 52%|█████▏    | 14/27 [00:04<00:03,  3.48it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.53it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.57it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.48it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.46it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.45it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.50it/s]
 78%|███████▊  | 21/27 [00:06<00:01,  3.49it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.50it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.50it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.50it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.49it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.49it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.41it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.36it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.36it/s]
 31%|███       | 4/13 [00:01<00:02,  3.40it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.43it/s]
 46%|████▌     | 6/13 [00:01<00:02,  3.44it/s]
 54%|█████▍    | 7/13 [00:02<00:01,  3.45it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.44it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.43it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.43it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.40it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.45it/s]
+
+
  0%|          | 0/3 [00:00<?, ?it/s]
 33%|███▎      | 1/3 [00:00<00:00,  3.07it/s]
 67%|██████▋   | 2/3 [00:00<00:00,  3.26it/s]
+
+
  0%|          | 0/8 [00:00<?, ?it/s]
 12%|█▎        | 1/8 [00:00<00:02,  3.48it/s]
 25%|██▌       | 2/8 [00:00<00:01,  3.43it/s]
 38%|███▊      | 3/8 [00:00<00:01,  3.47it/s]
 50%|█████     | 4/8 [00:01<00:01,  3.41it/s]
 62%|██████▎   | 5/8 [00:01<00:00,  3.45it/s]
 75%|███████▌  | 6/8 [00:01<00:00,  3.47it/s]
 88%|████████▊ | 7/8 [00:02<00:00,  3.47it/s]
+
+
  0%|          | 0/24 [00:00<?, ?it/s]
  4%|▍         | 1/24 [00:00<00:06,  3.39it/s]
  8%|▊         | 2/24 [00:00<00:06,  3.33it/s]
 12%|█▎        | 3/24 [00:00<00:06,  3.39it/s]
 17%|█▋        | 4/24 [00:01<00:05,  3.37it/s]
 21%|██        | 5/24 [00:01<00:05,  3.41it/s]
 25%|██▌       | 6/24 [00:01<00:05,  3.44it/s]
 29%|██▉       | 7/24 [00:02<00:04,  3.47it/s]
 33%|███▎      | 8/24 [00:02<00:04,  3.44it/s]
 38%|███▊      | 9/24 [00:02<00:04,  3.45it/s]
 42%|████▏     | 10/24 [00:02<00:04,  3.42it/s]
 46%|████▌     | 11/24 [00:03<00:03,  3.30it/s]
 50%|█████     | 12/24 [00:03<00:03,  3.32it/s]
 54%|█████▍    | 13/24 [00:03<00:03,  3.35it/s]
 58%|█████▊    | 14/24 [00:04<00:02,  3.35it/s]
 62%|██████▎   | 15/24 [00:04<00:02,  3.41it/s]
 67%|██████▋   | 16/24 [00:04<00:02,  3.43it/s]
 71%|███████   | 17/24 [00:05<00:02,  3.43it/s]
 75%|███████▌  | 18/24 [00:05<00:01,  3.40it/s]
 79%|███████▉  | 19/24 [00:05<00:01,  3.37it/s]
 83%|████████▎ | 20/24 [00:05<00:01,  3.34it/s]
 88%|████████▊ | 21/24 [00:06<00:00,  3.36it/s]
 92%|█████████▏| 22/24 [00:06<00:00,  3.39it/s]
 96%|█████████▌| 23/24 [00:06<00:00,  3.41it/s]
+
+
  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:02,  3.43it/s]
 22%|██▏       | 2/9 [00:00<00:02,  3.47it/s]
 33%|███▎      | 3/9 [00:00<00:01,  3.40it/s]
 44%|████▍     | 4/9 [00:01<00:01,  3.38it/s]
 56%|█████▌    | 5/9 [00:01<00:01,  3.44it/s]
 67%|██████▋   | 6/9 [00:01<00:00,  3.48it/s]
 78%|███████▊  | 7/9 [00:02<00:00,  3.53it/s]
 89%|████████▉ | 8/9 [00:02<00:00,  3.54it/s]
+
+
  0%|          | 0/27 [00:00<?, ?it/s]
  4%|▎         | 1/27 [00:00<00:07,  3.37it/s]
  7%|▋         | 2/27 [00:00<00:07,  3.49it/s]
 11%|█         | 3/27 [00:00<00:06,  3.53it/s]
 15%|█▍        | 4/27 [00:01<00:06,  3.55it/s]
 19%|█▊        | 5/27 [00:01<00:06,  3.53it/s]
 22%|██▏       | 6/27 [00:01<00:05,  3.58it/s]
 26%|██▌       | 7/27 [00:01<00:05,  3.53it/s]
 30%|██▉       | 8/27 [00:02<00:05,  3.52it/s]
 33%|███▎      | 9/27 [00:02<00:05,  3.54it/s]
 37%|███▋      | 10/27 [00:02<00:04,  3.54it/s]
 41%|████      | 11/27 [00:03<00:04,  3.53it/s]
 44%|████▍     | 12/27 [00:03<00:04,  3.52it/s]
 48%|████▊     | 13/27 [00:03<00:03,  3.51it/s]
 52%|█████▏    | 14/27 [00:03<00:03,  3.51it/s]
 56%|█████▌    | 15/27 [00:04<00:03,  3.49it/s]
 59%|█████▉    | 16/27 [00:04<00:03,  3.49it/s]
 63%|██████▎   | 17/27 [00:04<00:02,  3.51it/s]
 67%|██████▋   | 18/27 [00:05<00:02,  3.53it/s]
 70%|███████   | 19/27 [00:05<00:02,  3.53it/s]
 74%|███████▍  | 20/27 [00:05<00:01,  3.55it/s]
 78%|███████▊  | 21/27 [00:05<00:01,  3.54it/s]
 81%|████████▏ | 22/27 [00:06<00:01,  3.49it/s]
 85%|████████▌ | 23/27 [00:06<00:01,  3.50it/s]
 89%|████████▉ | 24/27 [00:06<00:00,  3.49it/s]
 93%|█████████▎| 25/27 [00:07<00:00,  3.45it/s]
 96%|█████████▋| 26/27 [00:07<00:00,  3.47it/s]
+
+
  0%|          | 0/12 [00:00<?, ?it/s]
  8%|▊         | 1/12 [00:00<00:03,  3.56it/s]
 17%|█▋        | 2/12 [00:00<00:02,  3.48it/s]
 25%|██▌       | 3/12 [00:00<00:02,  3.46it/s]
 33%|███▎      | 4/12 [00:01<00:02,  3.45it/s]
 42%|████▏     | 5/12 [00:01<00:01,  3.52it/s]
 50%|█████     | 6/12 [00:01<00:01,  3.54it/s]
 58%|█████▊    | 7/12 [00:01<00:01,  3.59it/s]
 67%|██████▋   | 8/12 [00:02<00:01,  3.58it/s]
 75%|███████▌  | 9/12 [00:02<00:00,  3.60it/s]
 83%|████████▎ | 10/12 [00:02<00:00,  3.58it/s]
 92%|█████████▏| 11/12 [00:03<00:00,  3.60it/s]
+
+
  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  3.67it/s]
 29%|██▊       | 2/7 [00:00<00:01,  3.67it/s]
 43%|████▎     | 3/7 [00:00<00:01,  3.61it/s]
 57%|█████▋    | 4/7 [00:01<00:00,  3.64it/s]
 71%|███████▏  | 5/7 [00:01<00:00,  3.70it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  3.63it/s]
+
+
  0%|          | 0/1 [00:00<?, ?it/s]
+
+
  0%|          | 0/13 [00:00<?, ?it/s]
  8%|▊         | 1/13 [00:00<00:03,  3.56it/s]
 15%|█▌        | 2/13 [00:00<00:03,  3.56it/s]
 23%|██▎       | 3/13 [00:00<00:02,  3.53it/s]
 31%|███       | 4/13 [00:01<00:02,  3.50it/s]
 38%|███▊      | 5/13 [00:01<00:02,  3.53it/s]
 46%|████▌     | 6/13 [00:01<00:01,  3.54it/s]
 54%|█████▍    | 7/13 [00:01<00:01,  3.53it/s]
 62%|██████▏   | 8/13 [00:02<00:01,  3.56it/s]
 69%|██████▉   | 9/13 [00:02<00:01,  3.57it/s]
 77%|███████▋  | 10/13 [00:02<00:00,  3.54it/s]
 85%|████████▍ | 11/13 [00:03<00:00,  3.57it/s]
 92%|█████████▏| 12/13 [00:03<00:00,  3.49it/s]
+
+
  0%|          | 0/20 [00:00<?, ?it/s]
  5%|▌         | 1/20 [00:00<00:05,  3.60it/s]
 10%|█         | 2/20 [00:00<00:04,  3.61it/s]
 15%|█▌        | 3/20 [00:00<00:04,  3.61it/s]
 20%|██        | 4/20 [00:01<00:04,  3.59it/s]
 25%|██▌       | 5/20 [00:01<00:04,  3.65it/s]
 30%|███       | 6/20 [00:01<00:03,  3.63it/s]
 35%|███▌      | 7/20 [00:01<00:03,  3.60it/s]
 40%|████      | 8/20 [00:02<00:03,  3.66it/s]
 45%|████▌     | 9/20 [00:02<00:02,  3.67it/s]
 50%|█████     | 10/20 [00:02<00:02,  3.62it/s]
 55%|█████▌    | 11/20 [00:03<00:02,  3.67it/s]
 60%|██████    | 12/20 [00:03<00:02,  3.64it/s]
 65%|██████▌   | 13/20 [00:03<00:01,  3.68it/s]
 70%|███████   | 14/20 [00:03<00:01,  3.69it/s]
 75%|███████▌  | 15/20 [00:04<00:01,  3.66it/s]
 80%|████████  | 16/20 [00:04<00:01,  3.70it/s]
 85%|████████▌ | 17/20 [00:04<00:00,  3.63it/s]
 90%|█████████ | 18/20 [00:04<00:00,  3.66it/s]
 95%|█████████▌| 19/20 [00:05<00:00,  3.63it/s]
+
+
  0%|          | 0/15 [00:00<?, ?it/s]
  7%|▋         | 1/15 [00:00<00:03,  3.59it/s]
 13%|█▎        | 2/15 [00:00<00:03,  3.70it/s]
 20%|██        | 3/15 [00:00<00:03,  3.62it/s]
 27%|██▋       | 4/15 [00:01<00:03,  3.63it/s]
 33%|███▎      | 5/15 [00:01<00:02,  3.60it/s]
 40%|████      | 6/15 [00:01<00:02,  3.66it/s]
 47%|████▋     | 7/15 [00:01<00:02,  3.55it/s]
 53%|█████▎    | 8/15 [00:02<00:01,  3.54it/s]
 60%|██████    | 9/15 [00:02<00:01,  3.51it/s]
 67%|██████▋   | 10/15 [00:02<00:01,  3.51it/s]
 73%|███████▎  | 11/15 [00:03<00:01,  3.52it/s]
 80%|████████  | 12/15 [00:03<00:00,  3.54it/s]
 87%|████████▋ | 13/15 [00:03<00:00,  3.58it/s]
 93%|█████████▎| 14/15 [00:03<00:00,  3.61it/s]
+
+[2025-07-11 17:23:31] Weighted Per image PCK0.10: 52.50%, image PCK0.05: 36.21%, image PCK0.01: 61.92%